Improving Machine Learning-Based Robot Self-Collision Checking with Input Positional Encoding

Bart{\l}omiej Kulecki; Dominik Belter

arxiv: 2509.07542 · v2 · submitted 2025-09-09 · 💻 cs.RO

Improving Machine Learning-Based Robot Self-Collision Checking with Input Positional Encoding

Bart{\l}omiej Kulecki , Dominik Belter This is my paper

Pith reviewed 2026-05-18 18:07 UTC · model grok-4.3

classification 💻 cs.RO

keywords robot self-collision detectionpositional encodingmachine learningmultilayer perceptroncollision checkingbinary classificationroboticshigh-frequency variations

0 comments

The pith

Positional encoding on input vectors raises accuracy of lightweight neural nets for robot self-collision checks.

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

The paper tests whether adding positional encoding to the raw joint-angle inputs of a small multilayer perceptron can improve binary classification of robot self-collisions. The encoding lets the network represent high-frequency changes in configuration space more faithfully, producing finer collision boundaries than an unencoded network of the same size. Because the resulting model stays lightweight, it evaluates faster than mesh-based geometric checks that rely on triangle intersections and bounding-volume hierarchies. If the improvement holds, motion planners could query collision status more often without sacrificing safety margins on articulated robots.

Core claim

The central claim is that feeding a positional encoding of the joint configuration vector into a lightweight multilayer perceptron enables the classifier to capture high-frequency variations that otherwise produce coarse or inaccurate collision decisions, thereby raising classification accuracy while retaining the speed advantage of learned models over classical geometric collision routines such as triangle-to-triangle tests and BVH queries.

What carries the argument

Positional encoding applied directly to the low-dimensional joint-configuration input of a binary-classification MLP.

If this is right

Machine-learning collision checkers can replace slower geometric routines inside real-time motion planners.
Small networks become viable for high-precision collision labeling once their inputs are positionally encoded.
Complex self-collision patterns in high-dimensional configuration spaces can be approximated without increasing model size.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The same input encoding might lift accuracy in other robot-learning tasks that involve continuous kinematic inputs, such as inverse kinematics or contact prediction.
Because the method adds no extra parameters at inference time, it could be dropped into existing deployed controllers with only a change in data preprocessing.
If the encoding generalizes across robot topologies, a single trained model could serve multiple arms without per-robot retraining.

Load-bearing premise

The lightweight MLP with positional encoding will continue to classify collision states correctly on joint configurations and limits it has never seen during training.

What would settle it

Measure classification accuracy and inference time on a held-out set of robot configurations drawn from different joint limits or a different robot morphology; if accuracy drops below the unencoded baseline or if geometric methods remain faster, the claimed benefit does not hold.

Figures

Figures reproduced from arXiv: 2509.07542 by Bart{\l}omiej Kulecki, Dominik Belter.

**Figure 2.** Figure 2: The configuration space divided into ranges with unique encoding values sign combina [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗

**Figure 3.** Figure 3: MLP architectures used in this research: MLP resulting from auto-sklearn optimization [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗

**Figure 4.** Figure 4: Influence of the number of training samples on the difference between training and testing [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗

**Figure 5.** Figure 5: Results of classification in 2D space: ground truth test data, predictions based on raw [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗

**Figure 6.** Figure 6: Mean classification accuracy for various positional encoding [PITH_FULL_IMAGE:figures/full_fig_p010_6.png] view at source ↗

**Figure 7.** Figure 7: Mean classification accuracy for various positional encoding [PITH_FULL_IMAGE:figures/full_fig_p011_7.png] view at source ↗

**Figure 8.** Figure 8: Mean classification accuracy for various positional encoding [PITH_FULL_IMAGE:figures/full_fig_p012_8.png] view at source ↗

**Figure 9.** Figure 9: Visualization of the predicted collision state in the robot workspace for the second [PITH_FULL_IMAGE:figures/full_fig_p013_9.png] view at source ↗

**Figure 10.** Figure 10: Average loss curves (based on 10 train trials) for training classifier in 2D space (x, y [PITH_FULL_IMAGE:figures/full_fig_p014_10.png] view at source ↗

**Figure 11.** Figure 11: Robot models used for collision checking in FCL and their wireframe views: precise [PITH_FULL_IMAGE:figures/full_fig_p015_11.png] view at source ↗

**Figure 12.** Figure 12: Mean inference time for various classifier architectures obtained using all tested input [PITH_FULL_IMAGE:figures/full_fig_p015_12.png] view at source ↗

read the original abstract

This manuscript investigates the integration of positional encoding -- a technique widely used in computer graphics -- into the input vector of a binary classification model for self-collision detection. The results demonstrate the benefits of incorporating positional encoding, which enhances classification accuracy by enabling the model to better capture high-frequency variations, leading to a more detailed and precise representation of complex collision patterns. The manuscript shows that machine learning-based techniques, such as lightweight multilayer perceptrons (MLPs) operating in a low-dimensional feature space, offer a faster alternative for collision checking than traditional methods that rely on geometric approaches, such as triangle-to-triangle intersection tests and Bounding Volume Hierarchies (BVH) for mesh-based models.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

Positional encoding gives a clear accuracy lift to the MLP on self-collision detection, but the paper does not show the model works outside the training joint ranges or on other robots.

read the letter

The main takeaway is that adding positional encoding to the joint inputs improves accuracy for a lightweight MLP doing robot self-collision classification, but we lack evidence that the improvement survives outside the training distribution. The paper applies a technique common in graphics and transformers to this robotics task. By encoding the input angles with sine and cosine functions at multiple frequencies, the model can represent sharper collision boundaries. They compare this against a plain MLP and report better classification performance while keeping the network small enough for fast inference. This is a practical step for motion planners that need to check collisions thousands of times per second, where geometric methods using BVH or triangle tests are too slow. The work does a reasonable job of framing the ML approach as a speed alternative to classic geometry. The idea is simple enough that others can replicate it quickly on their own robots. The main concern is whether the gains are real or just better fitting on the sampled data. The stress-test note is on point here: there is no demonstration that the model handles joint configurations beyond the training ranges or transfers to a different robot model. If the experiments stay within the same distribution, the accuracy boost might not reflect a deeper understanding of the collision manifold. The abstract also skips the actual numbers, dataset sizes, and ablation results, which makes it hard to judge the effect size. If the full manuscript includes those details and shows some cross-robot or out-of-range tests, the contribution strengthens. Otherwise it remains a narrow empirical tweak. This paper is aimed at robotics engineers working on real-time control and planning. Someone already using ML for collision avoidance would see immediate value in trying the encoding trick. It is not revolutionary, but it is the sort of targeted improvement that accumulates in the field. I would bring this to a reading group to talk about input features for robotics ML. It deserves serious peer review because the core experiment is straightforward and the problem is relevant, though the generalization claims need more support. I recommend sending it to referees with a request for additional validation on unseen configurations.

Referee Report

2 major / 2 minor

Summary. The manuscript investigates the integration of positional encoding into the input vector of a binary classification model for self-collision detection. The results demonstrate the benefits of incorporating positional encoding, which enhances classification accuracy by enabling the model to better capture high-frequency variations, leading to a more detailed and precise representation of complex collision patterns. The manuscript shows that machine learning-based techniques, such as lightweight multilayer perceptrons (MLPs) operating in a low-dimensional feature space, offer a faster alternative for collision checking than traditional methods that rely on geometric approaches, such as triangle-to-triangle intersection tests and Bounding Volume Hierarchies (BVH) for mesh-based models.

Significance. If the claimed accuracy improvements are substantiated through quantitative experiments with proper baselines and generalization tests, the work could offer a practical method for faster self-collision checking in robotics, reducing computational overhead compared to geometric methods while maintaining safety in motion planning.

major comments (2)

Abstract: The abstract asserts accuracy gains from positional encoding but provides no quantitative results, baselines, dataset details, or ablation controls, so the central claim cannot be verified from the available text.
Results/Evaluation: Generalization of the MLP+positional-encoding model to unseen joint configurations and robot models is not demonstrated. The central claim requires that adding positional encoding lets the model reliably detect self-collisions on configurations outside the training distribution, yet no such checks on joint values beyond sampled ranges, different kinematics, or changed joint limits are reported.

minor comments (2)

Provide explicit details on the positional encoding frequencies, MLP architecture (layer sizes, activations), training dataset size and sampling method, and exact accuracy metrics with comparisons.
Clarify the experimental setup, including how collision labels were generated and whether cross-validation or hold-out sets were used.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their thorough review and valuable comments on our manuscript. We address each of the major comments below and outline the revisions we plan to make.

read point-by-point responses

Referee: Abstract: The abstract asserts accuracy gains from positional encoding but provides no quantitative results, baselines, dataset details, or ablation controls, so the central claim cannot be verified from the available text.

Authors: We agree that the abstract could benefit from more specific information to allow readers to quickly assess the claims. The detailed quantitative results, including accuracy metrics, baseline comparisons (e.g., against standard MLPs and geometric methods), dataset descriptions, and ablation studies on positional encoding, are provided in the results section of the manuscript. In the revised version, we will expand the abstract to include key quantitative findings, such as the percentage improvement in accuracy and runtime reductions, while maintaining its conciseness. revision: yes
Referee: Results/Evaluation: Generalization of the MLP+positional-encoding model to unseen joint configurations and robot models is not demonstrated. The central claim requires that adding positional encoding lets the model reliably detect self-collisions on configurations outside the training distribution, yet no such checks on joint values beyond sampled ranges, different kinematics, or changed joint limits are reported.

Authors: We appreciate this observation regarding generalization. Our experiments were conducted on a specific robot model with joint configurations sampled across the operational range, demonstrating improved performance with positional encoding. However, we recognize that tests on explicitly out-of-distribution configurations, different robot models, or altered joint limits would further validate the approach. We will include additional experiments in the revised manuscript to evaluate generalization to unseen joint configurations and at least one additional robot model. revision: yes

Circularity Check

0 steps flagged

No circularity: purely empirical comparison with no derivations

full rationale

The manuscript is an empirical study comparing MLP-based binary classifiers for robot self-collision detection, with and without positional encoding on joint-angle inputs. No equations, first-principles derivations, or claimed predictions appear in the abstract or described results. Performance claims rest on measured classification accuracy on sampled configurations rather than any reduction of outputs to fitted parameters or self-citations by construction. The work is therefore self-contained against external benchmarks and receives the default non-circularity finding.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The approach rests on standard supervised learning assumptions plus the transfer of positional encoding from graphics to this robotics task; no new entities are postulated.

free parameters (1)

MLP network weights
All network parameters are fitted during training on collision-labeled pose data.

axioms (1)

domain assumption Positional encoding improves capture of high-frequency input variations for classification tasks
Invoked when stating that the technique enables better representation of complex collision patterns.

pith-pipeline@v0.9.0 · 5643 in / 1094 out tokens · 40728 ms · 2026-05-18T18:07:05.387826+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

26 extracted references · 26 canonical work pages

[1]

Efficient modeling and evaluation of constraints in path planning for multi- legged walking robots.IEEE Access, 7:107845–107862, 2019

Dominik Belter. Efficient modeling and evaluation of constraints in path planning for multi- legged walking robots.IEEE Access, 7:107845–107862, 2019

work page 2019
[2]

Learning-based proxy collision detection for robot motion plan- ning applications.IEEE Transactions on Robotics, 36(4):1096–1114, 2020

Nikhil Das and Michael Yip. Learning-based proxy collision detection for robot motion plan- ning applications.IEEE Transactions on Robotics, 36(4):1096–1114, 2020

work page 2020
[3]

Nikhil Das and Michael C. Yip. Forward kinematics kernel for improved proxy collision check- ing.IEEE Robotics and Automation Letters, 5(2):2349–2356, 2020

work page 2020
[4]

Eberhart

James Kennedy and Russell C. Eberhart. Particle swarm optimization. InProceedings of the IEEE International Conference on Neural Networks, pages 1942–1948, 1995

work page 1942
[5]

Fast kinodynamic planning on the constraint manifold with deep neural networks.IEEE Transactions on Robotics, 40:277–297, 2024

Piotr Kicki, Puze Liu, Davide Tateo, Haitham Bou-Ammar, Krzysztof Walas, Piotr Skrzypczy´ nski, and Jan Peters. Fast kinodynamic planning on the constraint manifold with deep neural networks.IEEE Transactions on Robotics, 40:277–297, 2024

work page 2024
[6]

Neural joint space implicit signed distance functions for reactive robot manipulator control.IEEE Robotics and Automation Letters, 8(2):480–487, 2022

Mikhail Koptev, Nadia Figueroa, and Aude Billard. Neural joint space implicit signed distance functions for reactive robot manipulator control.IEEE Robotics and Automation Letters, 8(2):480–487, 2022

work page 2022
[7]

Comparison of machine learning techniques for self-collisions checking of manipulating robots

Adam Krawczyk, Jakub Marciniak, and Dominik Belter. Comparison of machine learning techniques for self-collisions checking of manipulating robots. In2023 27th International Con- ference on Methods and Models in Automation and Robotics (MMAR), pages 472–477, 2023

work page 2023
[8]

Boosting machine learning techniques with positional encoding for robot collision checking,

Bart lomiej Kulecki and Dominik Belter. Boosting machine learning techniques with positional encoding for robot collision checking,. In13th International Workshop on Robot Motion and Control (RoMoCo), pages 90–95, 2024. 16

work page 2024
[9]

Positional encoding for robot neural self-collision checking

Bart lomiej Kulecki and Dominik Belter. Positional encoding for robot neural self-collision checking. InProceedings of the 5th Polish Conference on Artificial Intelligence, 2024

work page 2024
[10]

Doyle, Michael Guthe, and Jiˇ r´ ı Bittner

Daniel Meister, Shinji Ogaki, Carsten Benthin, Michael J. Doyle, Michael Guthe, and Jiˇ r´ ı Bittner. A survey on bounding volume hierarchies for ray tracing.Computer Graphics Forum, 40(2):683–712, 2023

work page 2023
[11]

Srinivasan, Matthew Tancik, Jonathan T

Ben Mildenhall, Pratul P. Srinivasan, Matthew Tancik, Jonathan T. Barron, Ravi Ramamoor- thi, and Ren Ng. NeRF: Representing scenes as neural radiance fields for view synthesis. In Andrea Vedaldi, Horst Bischof, Thomas Brox, and Jan-Michael Frahm, editors,Computer Vision – ECCV 2020, pages 405–421, Cham, 2020. Springer International Publishing

work page 2020
[12]

Differentiable collision detection: a randomized smoothing approach

Louis Montaut, Quentin Le Lidec, Antoine Bambade, Vladimir Petrik, Josef Sivic, and Justin Carpentier. Differentiable collision detection: a randomized smoothing approach. In2023 IEEE International Conference on Robotics and Automation (ICRA), pages 3240–3246, 2023

work page 2023
[13]

Instant neural graphics primitives with a multiresolution hash encoding.ACM Trans

Thomas M¨ uller, Alex Evans, Christoph Schied, and Alexander Keller. Instant neural graphics primitives with a multiresolution hash encoding.ACM Trans. Graph., 41(4):102:1–102:15, July 2022

work page 2022
[14]

FCL: A general purpose library for collision and proximity queries

Jia Pan, Sachin Chitta, and Dinesh Manocha. FCL: A general purpose library for collision and proximity queries. In2012 IEEE International Conference on Robotics and Automation, pages 3859–3866, 2012

work page 2012
[15]

Efficient probabilistic collision detection for non-gaussian noise distributions.IEEE Robotics and Automation Letters, 5(2):1024–1031, 2020

Jae Sung Park and Dinesh Manocha. Efficient probabilistic collision detection for non-gaussian noise distributions.IEEE Robotics and Automation Letters, 5(2):1024–1031, 2020

work page 2020
[16]

DeepSDF: Learning continuous signed distance functions for shape representation

Jeong Joon Park, Peter Florence, Julian Straub, Richard Newcombe, and Steven Lovegrove. DeepSDF: Learning continuous signed distance functions for shape representation. In2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 165– 174, 2019

work page 2019
[17]

Train short, test long: Attention with linear biases enables input length extrapolation

Ofir Press, Noah Smith, and Mike Lewis. Train short, test long: Attention with linear biases enables input length extrapolation. InInternational Conference on Learning Representations, 2022

work page 2022
[18]

Colin Raffel, Noam Shazeer, Adam Roberts, Katherine Lee, Sharan Narang, Michael Matena, Yanqi Zhou, Wei Li, and Peter J. Liu. Exploring the limits of transfer learning with a unified text-to-text transformer.J. Mach. Learn. Res., 21(1), January 2020

work page 2020
[19]

Ham- precht, Yoshua Bengio, and Aaron C

Nasim Rahaman, Aristide Baratin, Devansh Arpit, Felix Dr¨ axler, Min Lin, Fred A. Ham- precht, Yoshua Bengio, and Aaron C. Courville. On the spectral bias of neural networks. In International Conference on Machine Learning, 2018

work page 2018
[20]

Rosenblatt

F. Rosenblatt. The perceptron: A probabilistic model for information storage and organization in the brain.Psychological Review, 65(6):386–408, 1958

work page 1958
[21]

Leonard, and Luca Carlone

Antoni Rosinol, John J. Leonard, and Luca Carlone. NeRF-SLAM: Real-time dense monocular slam with neural radiance fields. In2023 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pages 3437–3444, 2023

work page 2023
[22]

Roformer: Enhanced transformer with rotary position embedding.Neurocomputing, 568:127063, 2024

Jianlin Su, Murtadha Ahmed, Yu Lu, Shengfeng Pan, Wen Bo, and Yunfeng Liu. Roformer: Enhanced transformer with rotary position embedding.Neurocomputing, 568:127063, 2024. 17

work page 2024
[23]

Srinivasan, Ben Mildenhall, Sara Fridovich-Keil, Nithin Raghavan, Utkarsh Singhal, Ravi Ramamoorthi, Jonathan T

Matthew Tancik, Pratul P. Srinivasan, Ben Mildenhall, Sara Fridovich-Keil, Nithin Raghavan, Utkarsh Singhal, Ravi Ramamoorthi, Jonathan T. Barron, and Ren Ng. Fourier features let networks learn high frequency functions in low dimensional domains. InProceedings of the 34th International Conference on Neural Information Processing Systems, NIPS’20, Red Hoo...

work page 2020
[24]

Howell, and Zachary Manchester

Kevin Tracy, Taylor A. Howell, and Zachary Manchester. Differentiable collision detection for a set of convex primitives. In2023 IEEE International Conference on Robotics and Automation (ICRA), pages 3663–3670, 2023

work page 2023
[25]

Caterpillar heuristic for gait-free planning with multi-legged robot.IEEE Robotics and Automation Letters, 8(8):5204–5211, 2023

David Valouch and Jan Faigl. Caterpillar heuristic for gait-free planning with multi-legged robot.IEEE Robotics and Automation Letters, 8(8):5204–5211, 2023

work page 2023
[26]

Gomez, Lukasz Kaiser, and Illia Polosukhin

Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Lukasz Kaiser, and Illia Polosukhin. Attention is all you need. InProceedings of the 31st In- ternational Conference on Neural Information Processing Systems, NIPS’17, page 6000–6010, Red Hook, NY, USA, 2017. Curran Associates Inc. 18

work page 2017

[1] [1]

Efficient modeling and evaluation of constraints in path planning for multi- legged walking robots.IEEE Access, 7:107845–107862, 2019

Dominik Belter. Efficient modeling and evaluation of constraints in path planning for multi- legged walking robots.IEEE Access, 7:107845–107862, 2019

work page 2019

[2] [2]

Learning-based proxy collision detection for robot motion plan- ning applications.IEEE Transactions on Robotics, 36(4):1096–1114, 2020

Nikhil Das and Michael Yip. Learning-based proxy collision detection for robot motion plan- ning applications.IEEE Transactions on Robotics, 36(4):1096–1114, 2020

work page 2020

[3] [3]

Nikhil Das and Michael C. Yip. Forward kinematics kernel for improved proxy collision check- ing.IEEE Robotics and Automation Letters, 5(2):2349–2356, 2020

work page 2020

[4] [4]

Eberhart

James Kennedy and Russell C. Eberhart. Particle swarm optimization. InProceedings of the IEEE International Conference on Neural Networks, pages 1942–1948, 1995

work page 1942

[5] [5]

Fast kinodynamic planning on the constraint manifold with deep neural networks.IEEE Transactions on Robotics, 40:277–297, 2024

Piotr Kicki, Puze Liu, Davide Tateo, Haitham Bou-Ammar, Krzysztof Walas, Piotr Skrzypczy´ nski, and Jan Peters. Fast kinodynamic planning on the constraint manifold with deep neural networks.IEEE Transactions on Robotics, 40:277–297, 2024

work page 2024

[6] [6]

Neural joint space implicit signed distance functions for reactive robot manipulator control.IEEE Robotics and Automation Letters, 8(2):480–487, 2022

Mikhail Koptev, Nadia Figueroa, and Aude Billard. Neural joint space implicit signed distance functions for reactive robot manipulator control.IEEE Robotics and Automation Letters, 8(2):480–487, 2022

work page 2022

[7] [7]

Comparison of machine learning techniques for self-collisions checking of manipulating robots

Adam Krawczyk, Jakub Marciniak, and Dominik Belter. Comparison of machine learning techniques for self-collisions checking of manipulating robots. In2023 27th International Con- ference on Methods and Models in Automation and Robotics (MMAR), pages 472–477, 2023

work page 2023

[8] [8]

Boosting machine learning techniques with positional encoding for robot collision checking,

Bart lomiej Kulecki and Dominik Belter. Boosting machine learning techniques with positional encoding for robot collision checking,. In13th International Workshop on Robot Motion and Control (RoMoCo), pages 90–95, 2024. 16

work page 2024

[9] [9]

Positional encoding for robot neural self-collision checking

Bart lomiej Kulecki and Dominik Belter. Positional encoding for robot neural self-collision checking. InProceedings of the 5th Polish Conference on Artificial Intelligence, 2024

work page 2024

[10] [10]

Doyle, Michael Guthe, and Jiˇ r´ ı Bittner

Daniel Meister, Shinji Ogaki, Carsten Benthin, Michael J. Doyle, Michael Guthe, and Jiˇ r´ ı Bittner. A survey on bounding volume hierarchies for ray tracing.Computer Graphics Forum, 40(2):683–712, 2023

work page 2023

[11] [11]

Srinivasan, Matthew Tancik, Jonathan T

Ben Mildenhall, Pratul P. Srinivasan, Matthew Tancik, Jonathan T. Barron, Ravi Ramamoor- thi, and Ren Ng. NeRF: Representing scenes as neural radiance fields for view synthesis. In Andrea Vedaldi, Horst Bischof, Thomas Brox, and Jan-Michael Frahm, editors,Computer Vision – ECCV 2020, pages 405–421, Cham, 2020. Springer International Publishing

work page 2020

[12] [12]

Differentiable collision detection: a randomized smoothing approach

Louis Montaut, Quentin Le Lidec, Antoine Bambade, Vladimir Petrik, Josef Sivic, and Justin Carpentier. Differentiable collision detection: a randomized smoothing approach. In2023 IEEE International Conference on Robotics and Automation (ICRA), pages 3240–3246, 2023

work page 2023

[13] [13]

Instant neural graphics primitives with a multiresolution hash encoding.ACM Trans

Thomas M¨ uller, Alex Evans, Christoph Schied, and Alexander Keller. Instant neural graphics primitives with a multiresolution hash encoding.ACM Trans. Graph., 41(4):102:1–102:15, July 2022

work page 2022

[14] [14]

FCL: A general purpose library for collision and proximity queries

Jia Pan, Sachin Chitta, and Dinesh Manocha. FCL: A general purpose library for collision and proximity queries. In2012 IEEE International Conference on Robotics and Automation, pages 3859–3866, 2012

work page 2012

[15] [15]

Efficient probabilistic collision detection for non-gaussian noise distributions.IEEE Robotics and Automation Letters, 5(2):1024–1031, 2020

Jae Sung Park and Dinesh Manocha. Efficient probabilistic collision detection for non-gaussian noise distributions.IEEE Robotics and Automation Letters, 5(2):1024–1031, 2020

work page 2020

[16] [16]

DeepSDF: Learning continuous signed distance functions for shape representation

Jeong Joon Park, Peter Florence, Julian Straub, Richard Newcombe, and Steven Lovegrove. DeepSDF: Learning continuous signed distance functions for shape representation. In2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 165– 174, 2019

work page 2019

[17] [17]

Train short, test long: Attention with linear biases enables input length extrapolation

Ofir Press, Noah Smith, and Mike Lewis. Train short, test long: Attention with linear biases enables input length extrapolation. InInternational Conference on Learning Representations, 2022

work page 2022

[18] [18]

Colin Raffel, Noam Shazeer, Adam Roberts, Katherine Lee, Sharan Narang, Michael Matena, Yanqi Zhou, Wei Li, and Peter J. Liu. Exploring the limits of transfer learning with a unified text-to-text transformer.J. Mach. Learn. Res., 21(1), January 2020

work page 2020

[19] [19]

Ham- precht, Yoshua Bengio, and Aaron C

Nasim Rahaman, Aristide Baratin, Devansh Arpit, Felix Dr¨ axler, Min Lin, Fred A. Ham- precht, Yoshua Bengio, and Aaron C. Courville. On the spectral bias of neural networks. In International Conference on Machine Learning, 2018

work page 2018

[20] [20]

Rosenblatt

F. Rosenblatt. The perceptron: A probabilistic model for information storage and organization in the brain.Psychological Review, 65(6):386–408, 1958

work page 1958

[21] [21]

Leonard, and Luca Carlone

Antoni Rosinol, John J. Leonard, and Luca Carlone. NeRF-SLAM: Real-time dense monocular slam with neural radiance fields. In2023 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pages 3437–3444, 2023

work page 2023

[22] [22]

Roformer: Enhanced transformer with rotary position embedding.Neurocomputing, 568:127063, 2024

Jianlin Su, Murtadha Ahmed, Yu Lu, Shengfeng Pan, Wen Bo, and Yunfeng Liu. Roformer: Enhanced transformer with rotary position embedding.Neurocomputing, 568:127063, 2024. 17

work page 2024

[23] [23]

Srinivasan, Ben Mildenhall, Sara Fridovich-Keil, Nithin Raghavan, Utkarsh Singhal, Ravi Ramamoorthi, Jonathan T

Matthew Tancik, Pratul P. Srinivasan, Ben Mildenhall, Sara Fridovich-Keil, Nithin Raghavan, Utkarsh Singhal, Ravi Ramamoorthi, Jonathan T. Barron, and Ren Ng. Fourier features let networks learn high frequency functions in low dimensional domains. InProceedings of the 34th International Conference on Neural Information Processing Systems, NIPS’20, Red Hoo...

work page 2020

[24] [24]

Howell, and Zachary Manchester

Kevin Tracy, Taylor A. Howell, and Zachary Manchester. Differentiable collision detection for a set of convex primitives. In2023 IEEE International Conference on Robotics and Automation (ICRA), pages 3663–3670, 2023

work page 2023

[25] [25]

Caterpillar heuristic for gait-free planning with multi-legged robot.IEEE Robotics and Automation Letters, 8(8):5204–5211, 2023

David Valouch and Jan Faigl. Caterpillar heuristic for gait-free planning with multi-legged robot.IEEE Robotics and Automation Letters, 8(8):5204–5211, 2023

work page 2023

[26] [26]

Gomez, Lukasz Kaiser, and Illia Polosukhin

Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Lukasz Kaiser, and Illia Polosukhin. Attention is all you need. InProceedings of the 31st In- ternational Conference on Neural Information Processing Systems, NIPS’17, page 6000–6010, Red Hook, NY, USA, 2017. Curran Associates Inc. 18

work page 2017