Pith. sign in

REVIEW 4 major objections 5 minor 103 references

Interaction-Merged Motion Planning: Effectively Leveraging Diverse Motion Datasets for Robust Planning

T0 review · 4 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read Merging per-module task vectors transfers interaction knowledge across motion datasets, beating ensembles and adaptation at zero extra inference cost.

desk verdict A well-motivated adaptation of model merging to motion planning with credible gains, but the 'diverse sources' claim needs a single-source ablation before it fully lands. read the letter →

arxiv 2507.04790 v3 pith:IVW2P7AK submitted 2025-07-07 cs.RO cs.AIcs.CVcs.LG

Giwon Lee , Wooseong Jeong , Daehee Park , Jaewoo Jeong , Kuk-Jin Yoon This is my paper
classification cs.ROcs.AIcs.CVcs.LG
keywords motionplanningmodelmergingtaskvectorcross-domainadaptationinteractionmodelingautonomousnavigationtrajectorydatasets
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper claims that an autonomous planner can absorb knowledge from many motion datasets without retraining on them all: take checkpoints trained separately on each source dataset, compute their task vectors, and learn per-module scalar weights that combine those vectors into one model for the target domain. On small target datasets such as SIT and THOR, this merged model beats training only on the target, domain adaptation, ensembling, and naive model-merging baselines on displacement, collision, and miss-rate metrics, while keeping inference cost at x1. The method also provides a stronger initialization: fine-tuning the merged model further improves every metric. The practical promise is a cheap, data-efficient route to cross-domain planning that reuses existing checkpoints and avoids domain imbalance and catastrophic forgetting.

What carries the argument

The load-bearing object is the task vector, defined as the difference between a trained checkpoint and an initial model. IMMP collects these vectors from source-domain checkpoints, groups parameters into ego, surrounding, interaction, and remaining decoder parameters, and optimizes a separate linear weight per group and per task vector against the target-domain loss. A second mechanism is the checkpoint pool: intermediate checkpoints at fixed intervals plus a best checkpoint for each metric, so the pool contains multiple snapshots of interaction patterns rather than a single fine-tuned model per domain.

What would settle it

Train the same planner on two source datasets, merge their task vectors with weights learned on half of a target dataset, and evaluate on the other half: if the merged model fails to beat a target-only baseline, or if re-running weight optimization on a different target split yields drastically different weights, the transfer claim is contradicted. A second counterexample would be a dataset pair where the best-per-metric checkpoints are mutually incompatible, causing catastrophic collision failure even though each alone performs well.

Watch

Extended reading notes

Core claim

The central discovery is that interaction-level information, not whole-model parameters, is what transfers across motion planning datasets. By partitioning a planner into ego-encoder, surrounding-encoder, interaction-encoder, and decoder groups, and merging each group's task vectors with its own learned scalar weights, IMMP preserves the feature hierarchy that connects agent behaviors to interactions. The checkpoint pool mixes intermediate checkpoints at fixed intervals with the best checkpoint per metric (ADE, FDE, collision rate, miss rate), so each source domain contributes multiple snapshots of interaction patterns. Empirical comparisons across GameTheoretic, DIPP, and DTPP backbones show that this interaction-level merge outperforms model-level merging, parameter-wise merging, ensembles, and domain adaptation on SIT and THOR targets.

Load-bearing premise

The load-bearing premise is that task vectors from independently trained planners can be linearly combined with learned weights, and that the module hierarchy (ego, surrounding, interaction, decoder) is a meaningful partition—a hierarchy the paper relaxes for DIPP by extracting weights at once from the unified planner-predictor module.

Editorial extensions

If this is right

  • Merged models reach target-domain performance without accessing source data during adaptation, keeping adaptation cheap and avoiding catastrophic forgetting.
  • The same merged parameters can serve as an initialization; fine-tuning them on the target set improves all metrics further.
  • Merging granularity matters: merging per interaction module outperforms whole-model or parameter-wise merging, suggesting planners should be merged along their feature hierarchy.
  • The method is backbone-agnostic: it composes with GameTheoretic, DIPP, and DTPP architectures without changing inference cost.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Beyond the paper: if linear mode connectivity holds for planning networks, the same pipeline could adapt one planner to many target domains by relearning only the scalar weights, without retraining source planners.
  • The learned merging weights appear proportional to domain similarity, so one could predict the weights from zero-shot target performance and skip the weight optimization entirely.
  • Because metric-specific checkpoints capture different interaction styles (collision-heavy versus miss-averse), the pool construction could be tuned toward a target's safety requirements rather than averaged across metrics.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The paper proposes Interaction-Merged Motion Planning (IMMP), a two-stage method for transferring knowledge from multiple source trajectory datasets into a target-domain motion planner. In the first stage, checkpoints are collected from models trained on each source domain, including metric-optimal checkpoints and intermediate checkpoints along the optimization trajectory. In the second stage, per-module task vectors are weighted and summed into an initial parameter vector, with the weights optimized on the target training set. The method is evaluated on SIT and THOR target domains with three planning backbones (GameTheoretic, DTPP, DIPP), reporting improved ADE, collision rate, FDE, and miss rate relative to domain generalization, domain adaptation, ensembles, and existing model-merging baselines at inference cost ×1. A supplementary study adds a Zara2 target and ablations on checkpoint composition, merging granularity, checkpoint interval, and module grouping.

Significance. If the reported results are robust, IMMP would be a practically attractive way to exploit existing motion datasets: it avoids joint multi-dataset training, adds no inference cost, and shows strong target-domain numbers across several backbones. The paper ships a code release, reports results on multiple planners and targets, and includes a useful supplementary analysis of checkpoint composition and hyperparameter sensitivity. However, the central transfer claim is currently underdetermined by the experiments: the learned merging weights could in principle collapse to selecting one useful source checkpoint, and the paper does not provide the single-source control needed to rule this out. In addition, the absence of repeated-seed variability, the target-dependent selection of checkpoint intervals, and the inconsistent treatment of DIPP's module granularity all limit confidence in the general claim. These are fixable with additional experiments and clarifications, so the work is suitable for major revision.

major comments (4)
  1. [Sec. 4.2, Table 1 and Algorithm 1] The central claim that IMMP leverages diverse source datasets is underdetermined without a single-source control. Because the merging weights w_{i,θ} in Eq. (1) are optimized on the target training set, the merged model can down-weight all but one source checkpoint. The paper never reports the target performance of a pool built from each source domain alone (or from random checkpoints of a single source) under the same weight-learning procedure; Fig. 3 is qualitative and no ablation removes source domains from the pool. If the best single-source pool already achieves the reported ADE, the claim that diverse interaction knowledge is merged rather than that one useful checkpoint is selected is unsupported. Please add quantitative single-source ablations and an oracle/upper-bound comparison.
  2. [Table 1 and Suppl. Table 6] The experimental comparison lacks repeated-seed variability, and the headline result appears to select the checkpoint interval C on the target domain. Table 1 reports single numbers with no error bars or seeds. Suppl. Table 6 sweeps C over 1, 2, 3, 10 and reports 'Ours (5)' as the selected value; there is no held-out protocol for choosing C, so the reported 0.3157 ADE may reflect target-set tuning. The table also labels this row as 'without finetuning' while the same number appears in Table 1 and Table 3 for IMMP+Finetune, which is internally inconsistent. Please report mean and standard deviation over at least three seeds and clarify the checkpoint-interval selection procedure.
  3. [Sec. 3.4 and Suppl. B] The method is claimed to be architecture-independent and to merge distinct modules (θego, θsurr, θinter, θelse), but this is not exercised on all three backbones. Suppl. B states that for DIPP 'weights are extracted at once from the unified module combining the planner and predictor.' Therefore the DIPP row in Table 1 does not test the proposed interaction-level grouping, and the claim of general applicability across planning models is weaker than presented. Please either report DIPP results with the same module-level separation, or clearly scope the claim to backbones where such separation is feasible.
  4. [Abstract and Sec. 5] The paper claims that IMMP 'mitigates domain imbalance and catastrophic forgetting,' but no experiment measures either phenomenon. The protocol collects source checkpoints once and does not sequentially update the model on source domains, so it avoids catastrophic forgetting by construction rather than mitigating it in a measurable sense; domain imbalance is also not directly evaluated. Please either add an evaluation that tracks source-domain performance after merging and fine-tuning, or soften the claims to avoid unsupported causal language.
minor comments (5)
  1. [Suppl. Table 8] The IMMP row in Table 8 reports FDE 0.9580 and Miss Rate 0.6976 for the same SIT/GameTheoretic setting, whereas Table 1 reports FDE 0.7626 and Miss Rate 0.6446; please reconcile the discrepancy.
  2. [Table 1] The row labeled 'Domain Generalization [17]' appears to describe the dataset-level approach of UniTraj, which is cited as [16] in the text; please align the citation labels.
  3. [Sec. 3.3] The statement that intermediate checkpoints 'tend to be more generalized across domain shifts' is an empirical claim that would benefit from a more direct analysis or citation, even though Table 3 provides supporting evidence.
  4. [Fig. 3 and Fig. 4] The caption states that w is 'the average of contributions of task vectors per epoch and per module,' but the exact definition is not given; please provide the formula or pseudo-code used to compute the displayed weights.
  5. [Eq. (6)-(9)] The metric definitions mix notation for ground-truth positions and planned positions (x and g); please define them uniformly and make clear that the collision threshold is applied to corresponding time steps of the plan and the ground truth.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the evaluation is held-out and the merging weights are fitted only to the target training split; the central claim is empirical rather than definitional.

full rationale

The paper's derivation chain is self-contained with respect to circularity. The task-vector formulation (Eq. 1) is taken from external prior work (Ilharco et al., Task Arithmetic), not derived from or justified by the target-domain results. The per-module merging weights w_{i,theta} are optimized on the target training set (Algorithm 1, lines 14-18) and all reported ADE, Collision Rate, FDE, and Miss Rate numbers are computed on held-out target validation or test splits (for SIT, the validation set is used because the test set lacks surrounding-agent ground truth, as stated in Suppl. A). Thus the reported improvements are not forced by construction: a learned linear combination of fixed source checkpoints could have failed on the evaluation split, and the baselines are trained or fine-tuned under comparable access to target data. Ablations in Tables 2, 3, 5, 6, and 7 compare design choices empirically rather than assuming their conclusions. The paper even honestly notes in Sec. 4.3 that the pre-merging phase is not strictly essential ('these results do not necessarily imply that the pre-merging phase is essential'), which further shows the claims are not definitionally tied to the method's components. Self-citations appear only in related-work context (e.g., refs. [24,25,26,27,49,50]) and are not load-bearing for any theorem or uniqueness claim. The absence of a single-source checkpoint-pool control is a limitation in evidence for the causal role of source diversity, but a missing control is an underdetermination concern, not circularity. No circular step could be identified and quoted from the paper.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The central claim rests on the empirical validity of linear model merging for planning networks and on a hand-specified module partition. The free parameters are hyperparameters, such as the checkpoint interval and learning rate, and the module grouping. No new physical entities are introduced.

free parameters (3)
  • checkpoint interval C = GameTheoretic forecaster: 30; GameTheoretic planner: 5; DTPP: 1; DIPP: 5
    Chosen per model; Table 6 sweeps C=1,2,3,5,10 on SIT and reports the best as 'Ours (5)'. This is a tuned hyperparameter, not a fixed rule.
  • merging weight optimizer learning rate = 1e-3 (GameTheoretic), 1e-2 (DTPP), 1e-3 (DIPP)
    Adam learning rate for updating the merging weights on the target train set; affects the final merged weights.
  • module grouping = {theta_ego, theta_surr, theta_inter, theta_else}
    Hand-chosen partition of the network into ego encoder, surrounding encoder, interaction encoder, and decoder; validated only on limited ablations.
assumptions (4)
  • domain assumption Task vectors from independently trained planners can be linearly combined into a usable merged model (linear mode connectivity).
    Eq. 1 in Sec. 3.2 assumes the merged theta is a weighted sum of checkpoints; the paper validates empirically but gives no proof that planning losses are linearly mode connected.
  • ad hoc to paper Intermediate checkpoints encode more transferable features than final checkpoints.
    Sec. 3.3 states the feature space of intermediate checkpoints 'tends to be more generalized across domain shifts'; no evidence is provided, and the paper later concedes the pre-merging phase is not essential.
  • domain assumption Planning models decompose into ego encoder, surrounding encoder, interaction encoder, and decoder, and these modules carry distinct transferable information.
    Sec. 3.4 groups parameters into theta_ego, theta_surr, theta_inter, and theta_else; this holds for the three backbones, but suppl B shows DIPP is merged differently, so the assumption is not universal.
  • domain assumption Naive joint training on multiple source datasets causes domain imbalance and catastrophic forgetting.
    Motivates the method in Sec. 1 and Sec. 3.1; the paper does not directly measure forgetting or imbalance in its own experiments.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Interaction-Merged Motion Planning: Effectively Leveraging Diverse Motion Datasets for Robust Planning." pith.science (2026). https://pith.science/paper/IVW2P7AK

@misc{pith2026250704790,
  author       = {Pith},
  title        = {Pith review of: Interaction-Merged Motion Planning: Effectively Leveraging Diverse Motion Datasets for Robust Planning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/IVW2P7AK}},
  note         = {Machine review of arXiv:2507.04790}
}
read the original abstract

Motion planning is a crucial component of autonomous robot driving. While various trajectory datasets exist, effectively utilizing them for a target domain remains challenging due to differences in agent interactions and environmental characteristics. Conventional approaches, such as domain adaptation or ensemble learning, leverage multiple source datasets but suffer from domain imbalance, catastrophic forgetting, and high computational costs. To address these challenges, we propose Interaction-Merged Motion Planning (IMMP), a novel approach that leverages parameter checkpoints trained on different domains during adaptation to the target domain. IMMP follows a two-step process: pre-merging to capture agent behaviors and interactions, sufficiently extracting diverse information from the source domain, followed by merging to construct an adaptable model that efficiently transfers diverse interactions to the target domain. Our method is evaluated on various planning benchmarks and models, demonstrating superior performance compared to conventional approaches.

Figures

Figures reproduced from arXiv: 2507.04790 by the authors.

Figure 1
Figure 1. Comparison of different approaches for leveraging the [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Overview of the proposed IMMP framework. In interaction-conserving pre-merging, parameter checkpoints are selected to [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Qualitative results on the SIT dataset with GameTheoretic planning model. (a) to (g) represent the inference results of SIT using [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Qualitative results on the THOR dataset with GameTheoretic planning model. From (a) to (g) represent the inference results for [PITH_FULL_IMAGE:figures/full_fig_p014_4.png]
Figure 5
Figure 5. Figure 5: Merging weights with respect to similarity to THOR. [PITH_FULL_IMAGE:figures/full_fig_p016_5.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

103 extracted references · 60 canonical work pages

  1. [1]

    Path planning of mobile robot with improved ant colony algo- rithm and mdp to produce smooth trajectory in grid-based environment

    Hub Ali, Dawei Gong, Meng Wang, and Xiaolin Dai. Path planning of mobile robot with improved ant colony algo- rithm and mdp to produce smooth trajectory in grid-based environment. Frontiers in neurorobotics, 14:44, 2020. 2

    2020

  2. [2]

    Ensemble of averages: Improving model selection and boosting performance in domain generalization

    Devansh Arpit, Huan Wang, Yingbo Zhou, and Caiming Xiong. Ensemble of averages: Improving model selection and boosting performance in domain generalization. Ad- vances in Neural Information Processing Systems, 35:8265– 8277, 2022. 3, 6, 7, 16

    2022

  3. [3]

    Use of relaxation methods in sampling-based algorithms for optimal mo- tion planning

    Oktay Arslan and Panagiotis Tsiotras. Use of relaxation methods in sampling-based algorithms for optimal mo- tion planning. In 2013 IEEE International Conference on Robotics and Automation, pages 2421–2428. IEEE, 2013. 2

    2013

  4. [4]

    Sit dataset: socially in- teractive pedestrian trajectory dataset for social navigation robots

    Jong Wook Bae, Jungho Kim, Junyong Yun, Changwon Kang, Jeongseon Choi, Chanhyeok Kim, Junho Lee, Jung- wook Choi, and Jun Won Choi. Sit dataset: socially in- teractive pedestrian trajectory dataset for social navigation robots. Advances in Neural Information Processing Systems, 36:24552–24563, 2023. 1, 6, 7, 13

    2023

  5. [5]

    Grid-based motion planning us- ing advanced motions for hexapod robots

    Wei Cheah, Hassan Hakim Khalili, Simon Watson, Peter Green, and Barry Lennox. Grid-based motion planning us- ing advanced motions for hexapod robots. In2018 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pages 3573–3578. IEEE, 2018. 2

    2018

  6. [6]

    Crowd-robot interaction: Crowd-aware robot navi- gation with attention-based deep reinforcement learning

    Changan Chen, Yuejiang Liu, Sven Kreiss, and Alexandre Alahi. Crowd-robot interaction: Crowd-aware robot navi- gation with attention-based deep reinforcement learning. In 2019 international conference on robotics and automation (ICRA), pages 6015–6022. IEEE, 2019. 1, 6, 13

    2019

  7. [7]

    End-to-end autonomous driving: Challenges and frontiers

    Li Chen, Penghao Wu, Kashyap Chitta, Bernhard Jaeger, An- dreas Geiger, and Hongyang Li. End-to-end autonomous driving: Challenges and frontiers. IEEE Transactions on Pat- tern Analysis and Machine Intelligence, 2024. 1

    2024

  8. [8]

    Ppad: Iterative interactions of prediction and planning for end-to-end autonomous driving

    Zhili Chen, Maosheng Ye, Shuangjie Xu, Tongyi Cao, and Qifeng Chen. Ppad: Iterative interactions of prediction and planning for end-to-end autonomous driving. In European Conference on Computer Vision , pages 239–256. Springer,

Show all 103 references
  1. [9]

    Forecast-mae: Self-supervised pre-training for motion forecasting with masked autoencoders

    Jie Cheng, Xiaodong Mei, and Ming Liu. Forecast-mae: Self-supervised pre-training for motion forecasting with masked autoencoders. In Proceedings of the IEEE/CVF In- ternational Conference on Computer Vision , pages 8679– 8689, 2023. 2

    2023

  2. [10]

    Fusing finetuned models for better pretraining

    Leshem Choshen, Elad Venezian, Noam Slonim, and Yoav Katz. Fusing finetuned models for better pretraining. arXiv preprint arXiv:2204.03044, 2022. 6, 7

    2022 arXiv

  3. [11]

    Adaptive stochastic weight averaging

    Caglar Demir, Arnab Sharma, and Axel-Cyrille Ngonga Ngomo. Adaptive stochastic weight averaging. arXiv preprint arXiv:2406.19092, 2024. 2, 3

    2024 arXiv

  4. [12]

    Imagenet: A large-scale hierarchical image database

    Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. Imagenet: A large-scale hierarchical image database. In 2009 IEEE conference on computer vision and pattern recognition, pages 248–255. Ieee, 2009. 4

    2009

  5. [13]

    Sparse instance conditioned multimodal trajectory predic- tion

    Yonghao Dong, Le Wang, Sanping Zhou, and Gang Hua. Sparse instance conditioned multimodal trajectory predic- tion. In Proceedings of the IEEE/CVF International Con- ference on Computer Vision, pages 9763–9772, 2023. 2

    2023

  6. [14]

    Recurrent aligned network for gen- eralized pedestrian trajectory prediction

    Yonghao Dong, Le Wang, Sanping Zhou, Gang Hua, and Changyin Sun. Recurrent aligned network for gen- eralized pedestrian trajectory prediction. arXiv preprint arXiv:2403.05810, 2024. 3

    2024 arXiv

  7. [15]

    An image is worth 16x16 words: Trans- formers for image recognition at scale

    Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Syl- vain Gelly, et al. An image is worth 16x16 words: Trans- formers for image recognition at scale. arXiv preprint a...

    2010 arXiv

  8. [16]

    Unitraj: A unified framework for scalable vehicle trajectory prediction

    Lan Feng, Mohammadhossein Bahari, Kaouther Mes- saoud Ben Amor, ´Eloi Zablocki, Matthieu Cord, and Alexan- dre Alahi. Unitraj: A unified framework for scalable vehicle trajectory prediction. In European Conference on Computer Vision, pages 106–123. Springer, 2024. 2, 3, 6, 7, 16

    2024

  9. [17]

    Uncertainty estimation for cross-dataset performance in trajectory prediction

    Thomas Gilles, Stefano Sabatini, Dzmitry Tsishkou, Bogdan Stanciulescu, and Fabien Moutarde. Uncertainty estimation for cross-dataset performance in trajectory prediction. arXiv preprint arXiv:2205.07310, 2022. 3, 6, 7, 16

    2022 arXiv

  10. [18]

    Planning-oriented autonomous driving

    Yihan Hu, Jiazhi Yang, Li Chen, Keyu Li, Chonghao Sima, Xizhou Zhu, Siqi Chai, Senyao Du, Tianwei Lin, Wenhai Wang, et al. Planning-oriented autonomous driving. In Pro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 17853–17862, 2023. 2

    2023

  11. [19]

    Emr-merging: Tuning-free high- performance model merging

    Chenyu Huang, Peng Ye, Tao Chen, Tong He, Xiangyu Yue, and Wanli Ouyang. Emr-merging: Tuning-free high- performance model merging. Advances in Neural Informa- tion Processing Systems, 37:122741–122769, 2024. 15, 16

    2024

  12. [20]

    Gameformer: Game-theoretic modeling and learning of transformer-based interactive prediction and planning for autonomous driving

    Zhiyu Huang, Haochen Liu, and Chen Lv. Gameformer: Game-theoretic modeling and learning of transformer-based interactive prediction and planning for autonomous driving. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 3903–3913, 2023. 1, 2

    2023

  13. [21]

    Dif- ferentiable integrated motion prediction and planning with learnable cost function for autonomous driving

    Zhiyu Huang, Haochen Liu, Jingda Wu, and Chen Lv. Dif- ferentiable integrated motion prediction and planning with learnable cost function for autonomous driving. IEEE trans- actions on neural networks and learning systems, 2023. 4, 6, 7, 13, 14

    2023

  14. [22]

    Dtpp: Differentiable joint conditional prediction and cost evaluation for tree policy planning in autonomous driving

    Zhiyu Huang, Peter Karkus, Boris Ivanovic, Yuxiao Chen, Marco Pavone, and Chen Lv. Dtpp: Differentiable joint conditional prediction and cost evaluation for tree policy planning in autonomous driving. In 2024 IEEE Inter- national Conference on Robotics and Automation (ICRA) , ...

    2024

  15. [23]

    Editing models with task arithmetic

    Gabriel Ilharco, Marco Tulio Ribeiro, Mitchell Wortsman, Suchin Gururangan, Ludwig Schmidt, Hannaneh Hajishirzi, and Ali Farhadi. Editing models with task arithmetic. arXiv preprint arXiv:2212.04089, 2022. 3, 4, 6, 7

    2022 arXiv

  16. [24]

    Multi-agent long-term 3d human pose forecasting via interaction-aware trajectory conditioning

    Jaewoo Jeong, Daehee Park, and Kuk-Jin Yoon. Multi-agent long-term 3d human pose forecasting via interaction-aware trajectory conditioning. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 1617–1628, 2024. 1

    2024

  17. [25]

    Multi-modal knowledge distillation-based 9 human trajectory forecasting

    Jaewoo Jeong, Seohee Lee, Daehee Park, Giwon Lee, and Kuk-Jin Yoon. Multi-modal knowledge distillation-based 9 human trajectory forecasting. In Proceedings of the Com- puter Vision and Pattern Recognition Conference , pages 24222–24233, 2025. 1

    2025

  18. [26]

    Quantifying task pri- ority for multi-task optimization

    Wooseong Jeong and Kuk-Jin Yoon. Quantifying task pri- ority for multi-task optimization. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 363–372, 2024. 3

    2024

  19. [27]

    Selective task group updates for multi-task optimization

    Wooseong Jeong and Kuk-Jin Yoon. Selective task group updates for multi-task optimization. arXiv preprint arXiv:2502.11986, 2025. 3

    2025 arXiv

  20. [28]

    Think twice be- fore driving: Towards scalable decoders for end-to-end au- tonomous driving

    Xiaosong Jia, Penghao Wu, Li Chen, Jiangwei Xie, Con- ghui He, Junchi Yan, and Hongyang Li. Think twice be- fore driving: Towards scalable decoders for end-to-end au- tonomous driving. In Proceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition, page...

    2023

  21. [29]

    Vad: Vectorized scene representa- tion for efficient autonomous driving

    Bo Jiang, Shaoyu Chen, Qing Xu, Bencheng Liao, Jiajie Chen, Helong Zhou, Qian Zhang, Wenyu Liu, Chang Huang, and Xinggang Wang. Vad: Vectorized scene representa- tion for efficient autonomous driving. In Proceedings of the IEEE/CVF International Conference on Computer Vision ,...

    2023

  22. [30]

    Fine-tuning linear layers only is a simple yet effective way for task arithmetic

    Ruochen Jin, Bojian Hou, Jiancong Xiao, Weijie Su, and Li Shen. Fine-tuning linear layers only is a simple yet effective way for task arithmetic. arXiv preprint arXiv:2407.07089 ,

    arXiv

  23. [31]

    Sampling-based algo- rithms for optimal motion planning

    Sertac Karaman and Emilio Frazzoli. Sampling-based algo- rithms for optimal motion planning. The international jour- nal of robotics research, 30(7):846–894, 2011. 2

    2011

  24. [32]

    Probabilistic roadmaps for path planning in high- dimensional configuration spaces

    Lydia E Kavraki, Petr Svestka, J-C Latombe, and Mark H Overmars. Probabilistic roadmaps for path planning in high- dimensional configuration spaces. IEEE transactions on Robotics and Automation, 12(4):566–580, 1996. 2

    1996

  25. [33]

    A game-theoretic framework for joint forecasting and planning

    Kushal Kedia, Prithwish Dan, and Sanjiban Choudhury. A game-theoretic framework for joint forecasting and planning. In 2023 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pages 6773–6778. IEEE, 2023. 1, 2, 4, 6, 7, 13, 14

    2023

  26. [34]

    Multi-task learning using uncertainty to weigh losses for scene geome- try and semantics

    Alex Kendall, Yarin Gal, and Roberto Cipolla. Multi-task learning using uncertainty to weigh losses for scene geome- try and semantics. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 7482–7491,

  27. [35]

    Overcoming catastrophic forgetting in neu- ral networks

    James Kirkpatrick, Razvan Pascanu, Neil Rabinowitz, Joel Veness, Guillaume Desjardins, Andrei A Rusu, Kieran Milan, John Quan, Tiago Ramalho, Agnieszka Grabska- Barwinska, et al. Overcoming catastrophic forgetting in neu- ral networks. Proceedings of the national academy of sc...

    2017

  28. [36]

    Rrt-connect: An efficient approach to single-query path planning

    James J Kuffner and Steven M LaValle. Rrt-connect: An efficient approach to single-query path planning. In Pro- ceedings 2000 ICRA. Millennium Conference. IEEE Inter- national Conference on Robotics and Automation. Symposia Proceedings (Cat. No. 00CH37065), pages 995–1001. IEEE,

    2000

  29. [37]

    Crowds by example

    Alon Lerner, Yiorgos Chrysanthou, and Dani Lischinski. Crowds by example. In Computer graphics forum , pages 655–664. Wiley Online Library, 2007. 1, 6, 13

    2007

  30. [38]

    An ensemble learning frame- work for vehicle trajectory prediction in interactive scenar- ios

    Zirui Li, Yunlong Lin, Cheng Gong, Xinwei Wang, Qi Liu, Jianwei Gong, and Chao Lu. An ensemble learning frame- work for vehicle trajectory prediction in interactive scenar- ios. In 2022 IEEE Intelligent Vehicles Symposium (IV), pages 51–57. IEEE, 2022. 3, 6

    2022

  31. [39]

    Conflict-averse gradient descent for multi-task learn- ing

    Bo Liu, Xingchao Liu, Xiaojie Jin, Peter Stone, and Qiang Liu. Conflict-averse gradient descent for multi-task learn- ing. Advances in Neural Information Processing Systems , 34:18878–18890, 2021. 3

    2021

  32. [40]

    Famo: Fast adaptive multitask optimization

    Bo Liu, Yihao Feng, Peter Stone, and Qiang Liu. Famo: Fast adaptive multitask optimization. Advances in Neural Information Processing Systems, 36:57226–57243, 2023. 3

    2023

  33. [41]

    Famo: Fast adaptive multitask optimization

    Bo Liu, Yihao Feng, Peter Stone, and Qiang Liu. Famo: Fast adaptive multitask optimization. Advances in Neural Information Processing Systems, 36, 2024. 3

    2024

  34. [42]

    Tangent transformers for composition, privacy and removal

    Tian Yu Liu, Aditya Golatkar, and Stefano Soatto. Tangent transformers for composition, privacy and removal. arXiv preprint arXiv:2307.08122, 2023. 3

    2023 arXiv

  35. [43]

    Jrdb: A dataset and bench- mark of egocentric robot visual perception of humans in built environments

    Roberto Martin-Martin, Mihir Patel, Hamid Rezatofighi, Abhijeet Shenoi, JunYoung Gwak, Eric Frankel, Amir Sadeghian, and Silvio Savarese. Jrdb: A dataset and bench- mark of egocentric robot visual perception of humans in built environments. IEEE transactions on pattern analysi...

    2021

  36. [44]

    Merging models with fisher-weighted averaging

    Michael S Matena and Colin A Raffel. Merging models with fisher-weighted averaging. Advances in Neural Information Processing Systems, 35:17703–17716, 2022. 4

    2022

  37. [45]

    Multi- task learning as a bargaining game

    Aviv Navon, Aviv Shamsian, Idan Achituve, Haggai Maron, Kenji Kawaguchi, Gal Chechik, and Ethan Fetaya. Multi- task learning as a bargaining game. arXiv preprint arXiv:2202.01017, 2022. 3

    2022 arXiv

  38. [46]

    Task arithmetic in the tangent space: Improved editing of pre-trained models

    Guillermo Ortiz-Jimenez, Alessandro Favero, and Pascal Frossard. Task arithmetic in the tangent space: Improved editing of pre-trained models. Advances in Neural Informa- tion Processing Systems, 36, 2024. 2, 3

    2024

  39. [47]

    Vlp: Vision language planning for autonomous driving

    Chenbin Pan, Burhaneddin Yaman, Tommaso Nesti, Abhirup Mallik, Alessandro G Allievi, Senem Velipasalar, and Liu Ren. Vlp: Vision language planning for autonomous driving. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 14760–14769, 2024. 2

    2024

  40. [48]

    Leveraging future relation- ship reasoning for vehicle trajectory prediction

    Daehee Park, Hobin Ryu, Yunseo Yang, Jegyeong Cho, Ji- won Kim, and Kuk-Jin Yoon. Leveraging future relation- ship reasoning for vehicle trajectory prediction. In The Eleventh International Conference on Learning Representa- tions, 2023. 2

    2023

  41. [49]

    Improv- ing transferability for cross-domain trajectory prediction via neural stochastic differential equation

    Daehee Park, Jaewoo Jeong, and Kuk-Jin Yoon. Improv- ing transferability for cross-domain trajectory prediction via neural stochastic differential equation. In Proceedings of the AAAI Conference on Artificial Intelligence , pages 10145– 10154, 2024. 3

    2024

  42. [50]

    T4p: Test-time training of tra- jectory prediction via masked autoencoder and actor-specific token memory

    Daehee Park, Jaeseok Jeong, Sung-Hoon Yoon, Jaewoo Jeong, and Kuk-Jin Yoon. T4p: Test-time training of tra- jectory prediction via masked autoencoder and actor-specific token memory. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 15...

    2024

  43. [51]

    You’ll never walk alone: Modeling social behav- ior for multi-target tracking

    Stefano Pellegrini, Andreas Ess, Konrad Schindler, and Luc Van Gool. You’ll never walk alone: Modeling social behav- ior for multi-target tracking. In2009 IEEE 12th international conference on computer vision, pages 261–268. IEEE, 2009. 1, 6, 13

    2009

  44. [52]

    Adaptraj: A multi-source domain generalization framework for multi-agent trajectory prediction

    Tangwen Qian, Yile Chen, Gao Cong, Yongjun Xu, and Fei Wang. Adaptraj: A multi-source domain generalization framework for multi-agent trajectory prediction. In 2024 IEEE 40th International Conference on Data Engineering (ICDE), pages 5048–5060. IEEE, 2024. 3

    2024

  45. [53]

    Th ¨or: Human-robot navigation data collection and accurate motion trajectories dataset

    Andrey Rudenko, Tomasz P Kucner, Chittaranjan S Swami- nathan, Ravi T Chadalavada, Kai O Arras, and Achim J Lilienthal. Th ¨or: Human-robot navigation data collection and accurate motion trajectories dataset. IEEE Robotics and Automation Letters, 5(2):676–682, 2020. 1, 6, 7, 13

    2020

  46. [54]

    Perceive, predict, and plan: Safe motion planning through interpretable seman- tic representations

    Abbas Sadat, Sergio Casas, Mengye Ren, Xinyu Wu, Pranaab Dhawan, and Raquel Urtasun. Perceive, predict, and plan: Safe motion planning through interpretable seman- tic representations. In Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Pro...

    2020

  47. [55]

    Navigation in human flows: planning with adaptive motion grid

    Jacques Saraydaryan, Fabrice Jumel, and Olivier Simonin. Navigation in human flows: planning with adaptive motion grid. In IROS Workshop CrowdNav, 2018. 2

    2018

  48. [56]

    Progress & compress: A scalable framework for continual learning

    Jonathan Schwarz, Wojciech Czarnecki, Jelena Luketina, Agnieszka Grabska-Barwinska, Yee Whye Teh, Razvan Pas- canu, and Raia Hadsell. Progress & compress: A scalable framework for continual learning. In International confer- ence on machine learning, pages 4528–4537. PMLR, 2018. 2

    2018

  49. [57]

    Independent component alignment for multi-task learning

    Dmitry Senushkin, Nikolay Patakin, Arseny Kuznetsov, and Anton Konushin. Independent component alignment for multi-task learning. In Proceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition, pages 20083–20093, 2023. 3

    2023

  50. [58]

    Continual learning with deep generative replay

    Hanul Shin, Jung Kwon Lee, Jaehong Kim, and Jiwon Kim. Continual learning with deep generative replay. Advances in neural information processing systems, 30, 2017. 2

    2017

  51. [59]

    Incremental learning of object detectors without catas- trophic forgetting

    Konstantin Shmelkov, Cordelia Schmid, and Karteek Ala- hari. Incremental learning of object detectors without catas- trophic forgetting. In Proceedings of the IEEE international conference on computer vision, pages 3400–3409, 2017. 2

    2017

  52. [60]

    Parameter efficient multi- task model fusion with partial linearization

    Anke Tang, Li Shen, Yong Luo, Yibing Zhan, Han Hu, Bo Du, Yixin Chen, and Dacheng Tao. Parameter efficient multi- task model fusion with partial linearization. arXiv preprint arXiv:2310.04742, 2023. 3

    2023 arXiv

  53. [61]

    Efficient evaluation of collisions and costs on grid maps for autonomous vehicle motion planning

    Georg Tanzmeister, Martin Friedl, Dirk Wollherr, and Mar- tin Buss. Efficient evaluation of collisions and costs on grid maps for autonomous vehicle motion planning. IEEE Trans- actions on Intelligent Transportation Systems , 15(5):2249– 2260, 2014. 2

    2014

  54. [62]

    Dreamwalker: Mental planning for contin- uous vision-language navigation

    Hanqing Wang, Wei Liang, Luc Van Gool, and Wen- guan Wang. Dreamwalker: Mental planning for contin- uous vision-language navigation. In Proceedings of the IEEE/CVF international conference on computer vision , pages 10873–10883, 2023. 1

    2023

  55. [63]

    Neural rrt*: Learning-based optimal path planning

    Jiankun Wang, Wenzheng Chi, Chenming Li, Chaoqun Wang, and Max Q-H Meng. Neural rrt*: Learning-based optimal path planning. IEEE Transactions on Automation Science and Engineering, 17(4):1748–1758, 2020. 2

    2020

  56. [64]

    Forecast-peft: Parameter- efficient fine-tuning for pre-trained motion forecasting mod- els

    Jifeng Wang, Kaouther Messaoud, Yuejiang Liu, Juergen Gall, and Alexandre Alahi. Forecast-peft: Parameter- efficient fine-tuning for pre-trained motion forecasting mod- els. arXiv preprint arXiv:2407.19564, 2024. 3

    2024 arXiv

  57. [65]

    Ganet: Goal area network for mo- tion forecasting

    Mingkun Wang, Xinge Zhu, Changqian Yu, Wei Li, Yuexin Ma, Ruochun Jin, Xiaoguang Ren, Dongchun Ren, Mingxu Wang, and Wenjing Yang. Ganet: Goal area network for mo- tion forecasting. In 2023 IEEE International Conference on Robotics and Automation (ICRA) , pages 1609–1615. IEEE,

    2023

  58. [66]

    Driving into the future: Multiview visual forecasting and planning with world model for au- tonomous driving

    Yuqi Wang, Jiawei He, Lue Fan, Hongxin Li, Yuntao Chen, and Zhaoxiang Zhang. Driving into the future: Multiview visual forecasting and planning with world model for au- tonomous driving. In Proceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition, pa...

    2024

  59. [67]

    Bridging the gap: Improving domain generalization in trajectory prediction

    Zhibo Wang, Jiayu Guo, Haiqiang Zhang, Ru Wan, Junping Zhang, and Jian Pu. Bridging the gap: Improving domain generalization in trajectory prediction. IEEE Transactions on Intelligent Vehicles, 2023. 3

    2023

  60. [68]

    Para-drive: Parallelized architecture for real- time autonomous driving

    Xinshuo Weng, Boris Ivanovic, Yan Wang, Yue Wang, and Marco Pavone. Para-drive: Parallelized architecture for real- time autonomous driving. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 15449–15458, 2024. 2

    2024

  61. [69]

    Model soups: averaging weights of multiple fine-tuned models improves accuracy without increasing in- ference time

    Mitchell Wortsman, Gabriel Ilharco, Samir Ya Gadre, Re- becca Roelofs, Raphael Gontijo-Lopes, Ari S Morcos, Hongseok Namkoong, Ali Farhadi, Yair Carmon, Simon Ko- rnblith, et al. Model soups: averaging weights of multiple fine-tuned models improves accuracy without increasing ...

    2022

  62. [70]

    Robust fine-tuning of zero-shot models

    Mitchell Wortsman, Gabriel Ilharco, Jong Wook Kim, Mike Li, Simon Kornblith, Rebecca Roelofs, Raphael Gon- tijo Lopes, Hannaneh Hajishirzi, Ali Farhadi, Hongseok Namkoong, et al. Robust fine-tuning of zero-shot models. In Proceedings of the IEEE/CVF conference on computer vi- ...

    2022

  63. [71]

    Adapting to length shift: Flexilength network for trajectory prediction

    Yi Xu and Yun Fu. Adapting to length shift: Flexilength network for trajectory prediction. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 15226–15237, 2024. 2

    2024

  64. [72]

    Adaptive trajectory prediction via transferable gnn

    Yi Xu, Lichen Wang, Yizhou Wang, and Yun Fu. Adaptive trajectory prediction via transferable gnn. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 6520–6531, 2022. 3

    2022

  65. [73]

    Training-free pretrained model merging

    Zhengqi Xu, Ke Yuan, Huiqiong Wang, Yong Wang, Mingli Song, and Jie Song. Training-free pretrained model merging. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 5915–5925, 2024. 15, 16

    2024

  66. [74]

    Ties-merging: Resolving interference 11 when merging models

    Prateek Yadav, Derek Tam, Leshem Choshen, Colin A Raf- fel, and Mohit Bansal. Ties-merging: Resolving interference 11 when merging models. Advances in Neural Information Pro- cessing Systems, 36:7093–7115, 2023. 2, 3, 4, 6, 7

    2023

  67. [75]

    Online learning for human classification in 3d lidar-based tracking

    Zhi Yan, Tom Duckett, and Nicola Bellotto. Online learning for human classification in 3d lidar-based tracking. In 2017 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pages 864–871. IEEE, 2017. 1

    2017

  68. [76]

    Diffusion-es: Gradient-free planning with diffusion for autonomous and instruction-guided driving

    Brian Yang, Huangyuan Su, Nikolaos Gkanatsios, Tsung- Wei Ke, Ayush Jain, Jeff Schneider, and Katerina Fragki- adaki. Diffusion-es: Gradient-free planning with diffusion for autonomous and instruction-guided driving. In Proceed- ings of the IEEE/CVF Conference on Computer Visi...

    2024

  69. [77]

    Adamerging: Adap- tive model merging for multi-task learning

    Enneng Yang, Zhenyi Wang, Li Shen, Shiwei Liu, Guibing Guo, Xingwei Wang, and Dacheng Tao. Adamerging: Adap- tive model merging for multi-task learning. arXiv preprint arXiv:2310.02575, 2023. 3, 4

    2023 arXiv

  70. [78]

    Model merging in llms, mllms, and beyond: Methods, theories, applications and opportunities

    Enneng Yang, Li Shen, Guibing Guo, Xingwei Wang, Xi- aochun Cao, Jie Zhang, and Dacheng Tao. Model merging in llms, mllms, and beyond: Methods, theories, applications and opportunities. arXiv preprint arXiv:2408.07666, 2024. 3

    2024 arXiv

  71. [79]

    Path- planning strategy for lane changing based on adaptive-grid risk-fields of autonomous vehicles

    Zhengcai Yang, Yunzhong Hu, and Youbing Zhang. Path- planning strategy for lane changing based on adaptive-grid risk-fields of autonomous vehicles. World Electric Vehicle Journal, 13(10):175, 2022. 2

    2022

  72. [80]

    Improv- ing the generalizability of trajectory prediction models with frenet-based domain normalization

    Luyao Ye, Zikang Zhou, and Jianping Wang. Improv- ing the generalizability of trajectory prediction models with frenet-based domain normalization. In 2023 IEEE Inter- national Conference on Robotics and Automation (ICRA) , pages 11562–11568. IEEE, 2023. 3

    2023

  73. [81]

    Gradient surgery for multi-task learning

    Tianhe Yu, Saurabh Kumar, Abhishek Gupta, Sergey Levine, Karol Hausman, and Chelsea Finn. Gradient surgery for multi-task learning. Advances in neural information process- ing systems, 33:5824–5836, 2020. 3

    2020

  74. [82]

    A survey of autonomous driving: Common practices and emerging technologies.IEEE access, 8:58443– 58469, 2020

    Ekim Yurtsever, Jacob Lambert, Alexander Carballo, and Kazuya Takeda. A survey of autonomous driving: Common practices and emerging technologies.IEEE access, 8:58443– 58469, 2020. 1

    2020

  75. [83]

    Dsdnet: Deep structured self-driving network

    Wenyuan Zeng, Shenlong Wang, Renjie Liao, Yun Chen, Bin Yang, and Raquel Urtasun. Dsdnet: Deep structured self-driving network. In Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part XXI 16, pages 156–172. Springer, 2020. 1

    2020

  76. [84]

    Genad: Generative end-to-end au- tonomous driving

    Wenzhao Zheng, Ruiqi Song, Xianda Guo, Chenming Zhang, and Long Chen. Genad: Generative end-to-end au- tonomous driving. In European Conference on Computer Vision, pages 87–104. Springer, 2024. 2

    2024

  77. [85]

    Hivt: Hierarchical vector transformer for multi-agent motion prediction

    Zikang Zhou, Luyao Ye, Jianping Wang, Kui Wu, and Ke- jie Lu. Hivt: Hierarchical vector transformer for multi-agent motion prediction. In Proceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition, pages 8823–8833, 2022. 2

    2022

  78. [86]

    Query-centric trajectory prediction

    Zikang Zhou, Jianping Wang, Yung-Hui Li, and Yu-Kai Huang. Query-centric trajectory prediction. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pat- tern Recognition, pages 17863–17873, 2023. 2

    2023

  79. [87]

    Avatargpt: All- in-one framework for motion understanding planning gener- ation and beyond

    Zixiang Zhou, Yu Wan, and Baoyuan Wang. Avatargpt: All- in-one framework for motion understanding planning gener- ation and beyond. In Proceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition, pages 1357–1366, 2024. 2

    2024

  80. [88]

    Unitraj: Universal human trajec- tory modeling from billion-scale worldwide traces

    Yuanshao Zhu, James Jianqiao Yu, Xiangyu Zhao, Xuetao Wei, and Yuxuan Liang. Unitraj: Universal human trajec- tory modeling from billion-scale worldwide traces. arXiv preprint arXiv:2411.03859, 2024. 2 12 Interaction-Merged Motion Planning: Effectively Leveraging Diverse Motio...

    2024

  81. [89]

    We alternately select 4 out of 5 scenes to form the train- ing and validation datasets, and train a separate model for each configuration

    Human-Human Interaction Dataset: ETH-UCY The ETH-UCY dataset [37, 51] consists of five sub- datasets: ETH, Hotel, Univ, Zara1, and Zara2, each with distinct pedestrian densities and scene characteristics. We alternately select 4 out of 5 scenes to form the train- ing and valid...

  82. [90]

    To model human interactions, the dataset first generates human movements by employing the ORCA algorithm, allowing agents to reach their desti- nations while avoiding collisions

    RL algorithm-based Robot Dataset: CrowdNav The CrowdNav dataset [6] is a simulation-based dataset designed to enable collision-free navigation in crowded environments. To model human interactions, the dataset first generates human movements by employing the ORCA algorithm, all...

  83. [91]

    The data was gath- ered in an indoor space measuring 8.4 × 18.8 m, with various fixed obstacles placed throughout

    Human-Robot Interaction Dataset: THOR and SIT The THOR dataset [53] is collected in an indoor environ- ment with real humans and a robot. The data was gath- ered in an indoor space measuring 8.4 × 18.8 m, with various fixed obstacles placed throughout. In this set- ting, real ...

  84. [92]

    ADE computes the L2 distance between every time step of the plan and the corresponding GT point, and then averages these dis- tances

    Average Displacement Error (ADE) ADE is a metric for evaluating effectiveness by assess- ing how similar the generated ego agent’s future plan is to the dataset’s ground truth trajectory. ADE computes the L2 distance between every time step of the plan and the corresponding GT...

  85. [93]

    It considers a collision to oc- cur when the distance between certain waypoints in the generated plan and the ground truth plan is below a spec- ified threshold

    Collision Rate (CR) Collision Rate is an important metric for evaluating safety in Motion Planning. It considers a collision to oc- cur when the distance between certain waypoints in the generated plan and the ground truth plan is below a spec- ified threshold. Following the a...

  86. [94]

    It calculates the L2 distance between the position at the final time step of the generated ego agent’s plan and the destina- tion

    Final Displacement Error (FDE) FDE is a metric for evaluating goal success. It calculates the L2 distance between the position at the final time step of the generated ego agent’s plan and the destina- tion. The formula for the FDE metric is as follows: FDE = 1 N NX i=1 ∥xTf i ...

  87. [95]

    We compute the L2 distance Table 4

    Miss Rate (MR) Miss rate is also a metric for evaluating goal success, assessing whether the position at the final time step of the generated plan deviates from the destination by more than a specified threshold. We compute the L2 distance Table 4. Checkpoint pool P constructe...

  88. [96]

    Specifically, we select the checkpoints where ADE, CR, FDE, and MR achieve their best values and store them in the checkpoint poolP

    Interaction-Conserving Pre-Merging We extract key checkpoints from each of the three trained planning baseline models. Specifically, we select the checkpoints where ADE, CR, FDE, and MR achieve their best values and store them in the checkpoint poolP. The selection of intermed...

  89. [97]

    Interaction Transfer with Merging Checkpoints in the checkpoint pool P are separated by module within the planning model. Specifically, we ex- tract weights from the LSTM layers responsible for em- bedding the trajectories of the ego agent and surrounding agents, denoted as θe...

  90. [98]

    As shown in Tab

    Need for Metric-wise Checkpoint Collection To demonstrate that metric-wise checkpoints capture distinct characteristics, we evaluated metric-specific checkpoints, selected on the Univ dataset, across two target domains. As shown in Tab. 5, checkpoints from the same source (Uni...

  91. [99]

    6, when the GameTheoretic model tar- gets the SIT domain, we evaluated performance across different checkpoint intervals C

    Hyperparameter Sensitivity: Checkpoint Interval As shown in Tab. 6, when the GameTheoretic model tar- gets the SIT domain, we evaluated performance across different checkpoint intervals C. Our method consis- tently outperforms the best-performing baseline in Tab. 1 (Ensemble-W...

  92. [100]

    A, the robot motion datasets differ in ego agent type and interaction mechanisms

    Ablation Study on Module Separation As shown in Sec. A, the robot motion datasets differ in ego agent type and interaction mechanisms. Our module separation is designed to reflect these characteristics. To validate its effectiveness, we conduct an ablation com- paring the orig...

  93. [101]

    As shown in Fig

    Correlation Between Domain Similarity and Merg- ing Weights We estimate the similarity between the source and tar- get domains by measuring zero-shot performance, and visualize the correlation between dataset similarity and the merging weights. As shown in Fig. 5, there exists...

  94. [102]

    8, evaluated on the SIT dataset using the GameThe- oretic planning model

    Comparison with Other Merging Techniques We compare with recent merging baselines listed in Tab. 8, evaluated on the SIT dataset using the GameThe- oretic planning model. Existing methods [19, 73] fail to preserve the hierarchical structure inherent in motion planning, resulti...

  95. [103]

    Such environments are likely to serve as realistic target domains in practical applications

    Experiments on a Larger Target Domain We selected SIT and THOR as target domains because they are real-world datasets collected in actual robotic navigation environments, where data collection is rela- tively challenging. Such environments are likely to serve as realistic targ...

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.