Robotics · arXiv:2610.18053 v1 · submitted 13 Oct 2026 · 23 references
LADDER: A Long-Horizon Robot Manipulation Benchmark
The paper says
“General-purpose robot manipulation policies learn short-horizon skills effectively, but degrade rapidly on long-horizon tasks. Existing benchmarks cannot isolate failures because task horizon is either too short or entangled with scene and object variations.”
more of the abstract
We propose LADDER, a long-horizon robot manipulation benchmark scaling from 1 to 10 subgoals within the same environment over three suites (SimplerEnv-Long, RobotArena-∞-Long, and RoboLab-Long), 145 tasks, eight categories, two embodiments, and 2,024 simulated episodes per system. We evaluate two orchestration paradigms: text subgoals for learned low-level policies (VoLo) and tool selection orchestration (CaP-X, GPT-Policy, PSP-Agent, TiPToP). Given the full instruction directly, low-level policies achieve 85.4% on single-subgoal tasks, but drop to near zero on extended tasks. While orchestration remedies this, it remains bounded by physical execution errors of the low-level policy rather than high-level planning and achieves only 25%. Tool selection orchestration with GPT-Policy on GPT-6 Astra significantly outperforms learned-policy methods, but incurs significant time and cost budget and succeeds only 62%. Real-world experiments across two embodiments confirm the collapse of the low-level policy and the gain from orchestration. Project website: https://ladder.insait.ai
from the abstract
Pith says
Machine review · deepseek-v4-flash · 14 Oct 2026 · what was read was not recorded · not checked by a person
It can be wrong. Challenge this review or reply to it.
Holds up. LADDER is a well-designed benchmark that cleanly isolates task horizon; the evaluation is thorough and the limitations are honestly stated. Worth engaging with. This line is wrong
6 minor
How often objections from reviews like this one are right: not yet measured
Not checked: figures, code, what the 23 references say.
6 minor objections
The paper says, at Section 1 (Introduction) – contributions list
Pith says: minor objection 1 of 6
The three contributions in the bullet list each run for several lines and include multiple results. Breaking each into a short statement followed by a separate sentence of elaboration would improve readability.
The paper says, at Figure 4 (Performance scaling)
Pith says: minor objection 2 of 6
The legend lists systems top to bottom, but the lines in the plot use the same colors in an order that does not match the legend exactly (e.g., green for Proprioceptive verifier is below yellow for VoLo). Adding a marker for each line or reordering the legend to match the plot would reduce ambiguity.
show 4 more
The paper says, at Table 3 (Task success rate) – formatting
Pith says: minor objection 3 of 6
The table uses footnotes to indicate the best learned-policy method (green) and best tool-selection method (blue), but the cells are not actually colored in this preprint. If the journal allows, applying color directly to the winning entries per suite would make the comparison instantaneous.
The paper says, at §5.2 (Failure analysis) and Appendix H
Pith says: minor objection 4 of 6
The measurement precision caveat in Appendix H (73.0% agreement for placement classification) is important but only appears in the appendix. A brief note in the main text (e.g., after the failure cause breakdown) would prevent overconfidence in the `slightly off target` category.
The paper says, at §5.3 (Real-world experiments)
Pith says: minor objection 5 of 6
The real-world evaluation uses only 12 episodes per task and system (3 layouts × 4 trials), which gives wide confidence intervals. The paper states "preserved rankings" but should explicitly quantify the uncertainty (e.g., Wilson intervals for real-world results are absent from Table 10 except for the combined rows).
The paper says, at §5 (Experimental setup) – depth usage
Pith says: minor objection 6 of 6
The paper acknowledges that tool-selection systems use ground-truth depth in simulation, but this is stated only in the middle of a paragraph. Adding a sentence in the conclusions to reiterate that this gives an advantage not solely attributable to orchestration would prevent overgeneralization.
What would settle it
As a sanity check, reproduce the main horizon-scaling curve (Figure 4) with a different low-level policy (e.g., RT-2 or a diffusion policy from the Open X-Embodiment dataset) on the SimplerEnv-Long suite. If the collapse pattern is qualitatively similar, the finding is robust to policy choice. If not, the claim is specific to the tested policy family.
from the skeptic's reading
How the review went
4 readings by deepseek-v4-flash · 14 Oct 2026
- Reader holds up
- Skeptic agrees
- Editor wrote the verdict above
- Referee minor revision: 6 minor
Not run on this paper: the simulated authors' defence.
Pith did not check this review's objections against the paper.
The review in full
The review in brief, machine-written
LADDER fills a real gap. Existing long-horizon benchmarks either keep tasks short or confound horizon with scene variation and object difficulty. LADDER solves this by building incremental ladders: tasks that add subgoals one at a time while keeping the environment and grasp difficulty fixed. That is genuinely useful for diagnosing where long-horizon systems fail. The paper does a thorough job of evaluating ten systems (including their own PSP-Agent) across three suites, two embodiments, 145 tasks, and 2,024 simulated episodes per system. They report 95% Wilson intervals, run paired bootstrap tests, and include real-robot validation (though limited). The failure analysis is detailed and separates orchestration mistakes from physical execution errors. The main findings—low-level policies collapse on multi-subgoal tasks; orchestration helps but is bounded by execution errors; tool-selection methods like GPT-Policy perform best but at high cost—are well-supported.
Soft spots: (1) Sim-to-real correlation is load-bearing but the real-robot experiments cover only horizons 1–5 with 12 episodes per task. The authors acknowledge this, and the trend matches simulation, but the quantitative rankings might shift with more real data. (2) The benchmark is restricted to tabletop pick-and-place; it will not test pushing, sliding, or articulated objects. That is a scope limitation, clearly stated. (3) The best tool-selection systems (GPT-Policy, PSP-Agent) use ground-truth depth in simulation, while learned-policy methods get only camera images. The authors note this, but it partly explains the gap. (4) GPT-Policy costs $4 per simulated episode ($8,134 for a full run); the paper is honest about this, but the benchmark’s usefulness for routine evaluation depends on cheaper proxies. (5) The offline VLM benchmark (Table 4) is interesting but only weakly predictive of online success: VLMs decompose instructions at 93% but execution still fails.
Overall this is a solid, reproducible, and clearly scoped benchmark. The limitations are openly listed and do not undermine the core contribution. I would send this to peer review with an invitation to address the sim-to-real generalisation more explicitly (e.g., by providing a protocol for others to run real-robot validation).
The strongest objection
The central claim that task horizon is a primary failure mode for generalist robot manipulation policies is well-supported by the experimental data. The benchmark design is rigorous, the analysis is thorough, and the limitations (limited real-robot coverage, specific low-level policies, tabletop pick-and-place only) are explicitly stated and do not undermine the core conclusion. The paper provides sufficient evidence that LADDER is a useful and methodologically sound benchmark for evaluating long-horizon manipulation.
The reader's verdict of ACCEPT is appropriate. The paper makes a clear, well-evidenced contribution. The sim-to-real assumption is partially supported by prior work and limited real-robot experiments, and the paper acknowledges its limitations. No additional load-bearing concern emerged from the stress-test; the internal consistency and methodological rigor are high.
The referee's report
The benchmark is carefully designed to isolate task horizon as a variable, with 138 of 145 tasks being incremental (a longer task adds subgoals to a shorter one), programmatic per-subgoal predicates, and coverage of eight spatial/logical task categories. The experimental methodology is rigorous: 95% Wilson confidence intervals, paired bootstrap tests across tasks, and detailed failure-cause analysis. The finding that task horizon is a primary failure mode for generalist manipulation policies, and that physical execution—not high-level planning—is the dominant bottleneck, is clearly supported by the data. The real-world validation, though limited in scale, shows consistent ranking with simulation. LADDER will be a valuable resource for the community studying long-horizon manipulation.
Pith's reading of the paper (a machine's summary, not the authors')
The paper establishes that task horizon is a primary failure mode for generalist robot manipulation policies. The authors introduce LADDER, a benchmark of 145 pick-and-place tasks that systematically scales from 1 to 10 subgoals while holding the environment and object dimensions fixed. Low-level policies achieve 85.4% success on single-subgoal tasks but drop to near zero when given five or more subgoals directly. Hierarchical orchestration—decomposing instructions into subgoals, visually verifying completion, and retrying failures—roughly triples multi-subgoal success but still tops out at about 25% for learned-policy backends. Tool-selection methods that call motion primitives instead of a learned policy reach 62.0% success (GPT-Policy) but at impractical cost, while a simpler tool-selection agent (PSP-Agent) reaches 40.1% using the same vision-language model. Physical execution errors, not high-level planning, are the main bottleneck.
What carries the argument. The central object is the LADDER benchmark itself, which provides controlled horizon scaling by building 'ladders' of tasks: within a fixed scene and task category, longer tasks retain all subgoals of shorter ones and add more. This isolates task horizon from other confounding factors. The benchmark also provides programmatic per-subgoal success predicates, covers eight task categories (single destination, split destination, transfer, relay, revisit, exclusion, ordered, duplicates), and includes 138 incremental subgoal tasks out of 145. The paper's analysis framework characterizes orchestrators along four dimensions: subgoal planning (text, code, or ground truth), trigger mechanism (periodic, proprioceptive, privileged, or primitive completion), visual verification (none, VLM-based, geometric, or simulator state), and failure recovery (advance, retry, replan). These dimensions are used to decompose and compare ten systems across learned-policy and tool-selection paradigms.
Core claim. The central discovery is that task horizon, disentangled from scene clutter and instruction phrasing, causes learned low-level manipulation policies to fail catastrophically. Using the LADDER benchmark—three simulation suites based on existing platforms, 145 tasks, eight spatial and logical categories, and 2,024 episodes per system—the paper shows that a π0-like policy trained on BridgeData v2 drops from 85.4% on single-subgoal tasks to 0.0% at five subgoals when given the full instruction directly. Orchestration methods that plan subgoals and verify completion recover performance to roughly 20–25% on average, but are still limited by the low-level policy's inability to execute the required pick-and-place actions reliably. Tool-selection backends (PSP-Agent, GPT-Policy) that bypass the learned policy achieve higher success (40–62%), confirming that the low-level policy's execution errors, not high-level reasoning, are the bottleneck. The benchmark is validated with real-world experiments on two robot platforms, which preserve the ranking of methods seen in simulation.
The premise it rests on. The quantitative rankings produced by LADDER in simulation are assumed to predict real-world performance, based on prior evidence that the underlying simulators correlate with real setups, while the paper's own real-robot experiments cover only a limited subset (1–5 subgoals, 12 episodes per task) and the paper acknowledges that visual state tracking remains challenging under heavy clutter.
If the paper is right (Pith's inferences, not the authors')
Future research should prioritize improving low-level policy robustness for long-horizon execution, since even ground-truth subgoal instructions still fail due to physical execution errors.
Tool-selection orchestration (calling motion primitives rather than a learned policy) substantially outperforms learned-policy backends, suggesting that decoupling planning from execution is a promising direction.
The LADDER benchmark provides a standardized way to measure long-horizon manipulation capability, enabling fair comparison across systems and tracking progress over time.
VLM-based visual verification, while helpful, still misses many completion states (recall ~62%) and remains a weak link; better state tracking methods are needed.
The high cost of GPT-Policy ($6.49 per successful episode) suggests that practical deployment of long-horizon robot manipulation may require cheaper alternatives, perhaps combining simpler verification with robust low-level primitives.
The dominant failure mode for learned-policy systems—'subgoals never started'—indicates that policies get stuck on early subgoals and exhaust the time budget; better recovery strategies that detect stalling and reset or re-trigger could help.
The benchmark's simulated results preserve method rankings on real robots for the subset tested, but the real-robot evaluation covers only 1–5 subgoals and 12 episodes per task; confirming the collapse pattern on more diverse real-world scenes would strengthen the claim.
Because the privileged verifier (access to ground-truth subgoal completion) still achieves only 29.6% overall success, even perfect state tracking and planning leave most of the performance gap, pointing to the low-level policy's inability to reliably execute individual subgoals as the core challenge.
Check this review
Pith stored no record of this review's passes and took no fingerprint of it. Reviews made by code from 13 Nov 2026 on carry both.
How often reviews like this one are right
Reviews like this one: written by deepseek-v4-flash, where what was read was not recorded.
How often objections from reviews like this one are right: not yet measured
Reviews by people
This review was written by a machine. No person has reviewed this paper on Pith yet. A scientist listed on Pith can review it for a fee, and their review is published here whatever it concludes.
The paper
Abstract
General-purpose robot manipulation policies learn short-horizon skills effectively, but degrade rapidly on long-horizon tasks. Existing benchmarks cannot isolate failures because task horizon is either too short or entangled with scene and object variations. We propose LADDER, a long-horizon robot manipulation benchmark scaling from 1 to 10 subgoals within the same environment over three suites (SimplerEnv-Long, RobotArena-∞-Long, and RoboLab-Long), 145 tasks, eight categories, two embodiments, and 2,024 simulated episodes per system. We evaluate two orchestration paradigms: text subgoals for learned low-level policies (VoLo) and tool selection orchestration (CaP-X, GPT-Policy, PSP-Agent, TiPToP). Given the full instruction directly, low-level policies achieve 85.4% on single-subgoal tasks, but drop to near zero on extended tasks. While orchestration remedies this, it remains bounded by physical execution errors of the low-level policy rather than high-level planning and achieves only 25%. Tool selection orchestration with GPT-Policy on GPT-6 Astra significantly outperforms learned-policy methods, but incurs significant time and cost budget and succeeds only 62%. Real-world experiments across two embodiments confirm the collapse of the low-level policy and the gain from orchestration. Project website: https://ladder.insait.ai
The paper's own sentences, arranged, kept beside the review and separate from it. Each label is drawn from a closed vocabulary; the number beside it is the probability the classifier gave that label. No sentence in this record was written by a model.
abstract compiled · 2026-10-14
The authors' main claim, as the classifier picked it out 0.69
“While orchestration remedies this, it remains bounded by physical execution errors of the low-level policy rather than high-level planning and achieves only ~25%.”
How the main claim is stated: asserted 1.00
What the abstract itself states, as probabilities: claims first 0.13 · states limitation 0.86 · human participants 0.04 · controlled comparison 0.91 · randomized assignment 0.02 · states quantitative result 0.99 · compares with prior results 0.92 · mentions code or data availability 0.35
The abstract, sentence by sentence
This table labels the abstract only (arXiv listing abstract v1). The number is the probability the classifier gave the label.
| Sentence | Role | The authors' own work | States the main conclusion | Strength |
|---|---|---|---|---|
| General-purpose robot manipulation policies learn short-horizon skills effectively, but degrade rapidly on long-horizon tasks. | background 0.99 | 0.13 | 0.06 | asserted 1.00 |
| Existing benchmarks cannot isolate failures because task horizon is either too short or entangled with scene and object variations. | gap 0.98 | 0.08 | 0.03 | asserted 1.00 |
| We propose LADDER, a long-horizon robot manipulation benchmark scaling from 1 to 10 subgoals within the same environment over three suites (SimplerEnv-Long, RobotArena-$\infty$-Long, and RoboLab-Long), 145 tasks, eight categories, two embodiments, and 2,024 simulated episodes per system. | contribution 0.87 | 0.94 | 0.06 | asserted 0.99 |
| We evaluate two orchestration paradigms: text subgoals for learned low-level policies (VoLo) and tool selection orchestration (CaP-X, GPT-Policy, PSP-Agent, TiPToP). | method 0.84 | 0.91 | 0.04 | asserted 0.94 |
| Given the full instruction directly, low-level policies achieve 85.4% on single-subgoal tasks, but drop to near zero on extended tasks. | result 1.00 | 0.86 | 0.13 | asserted 1.00 |
| While orchestration remedies this, it remains bounded by physical execution errors of the low-level policy rather than high-level planning and achieves only ~25%. | result 0.75 | 0.85 | 0.28 | asserted 1.00 |
| Tool selection orchestration with GPT-Policy on GPT-6 Astra significantly outperforms learned-policy methods, but incurs significant time and cost budget and succeeds only ~62%. | result 1.00 | 0.88 | 0.18 | asserted 1.00 |
| Real-world experiments across two embodiments confirm the collapse of the low-level policy and the gain from orchestration. | result 0.93 | 0.94 | 0.20 | asserted 1.00 |
| Project website: https://ladder.insait.ai | contribution 0.43 | 0.13 | 0.02 | unclear 0.98 |
What kind of paper the classifier reports this is: dataset or benchmark 1.00 · evidence from observation or experiment 0.65
What was examined, and what was not
Sources read
- arXiv listing abstract v1
Not examined
- full text
- figures and tables
- references and citing sentences
- relations to other papers
Levels run
- P abstract
Not run
- S sections
- T body sentences
- C citations
- R cross-paper relations
How the labels were made
Compiled by classifier-1.13.0 via TypeSafe API, question bank abstract-p1-20260923, 2026-10-14: 1 request, 7,014 input tokens.
Every quoted sentence on this record is the authors'. Every label is from Pith's closed vocabulary, with the probability shown beside it. This record as JSON
Figures (8)
Figures are the authors', reproduced from the paper.
References (23)
- [1] Michael Ahn, Anthony Brohan, Noah Brown, Yevgen Chebotar, Omar Cortes, Byron David, Chelsea Finn, Chuyuan Fu, Keerthana Gopalakrishnan, Karol Hausman, Alex Herzog, Daniel Ho, Jasmine Hsu, Julian Ibarz, Brian Ichter, Alex Irpan, Eric Jang, Rosario Jauregui Ruano, Kyle Jeffrey, ... on Pith
- [2] Shuai Bai, Yuxuan Cai, Ruizhe Chen, Keqin Chen, Xionghui Chen, Zesen Cheng, Lianghao Deng, Wei Ding, Chang Gao, Chunjiang Ge, Wenbin Ge, Zhifang Guo, Qidong Huang, Jie Huang, Fei Huang, Binyuan Hui, Shutong Jiang, Zhaohai Li, Mingsheng Li, Mei Li, Kaixin Li, Zicheng Lin, Junya... on PithPith's reviewarXiv
- [3] Suneel Belkhale, Tianli Ding, Ted Xiao, Pierre Sermanet, Quon Vuong, Jonathan Tompson, Yevgen Chebotar, Debidatta Dwibedi, and Dorsa Sadigh. RT-H : Action hierarchies using language. In Robotics: Science and Systems (RSS), 2024 on Pith
- [4] Kevin Black, Noah Brown, James Darpinian, Karan Dhabalia, Danny Driess, Adnan Esmail, Michael Robert Equi, Chelsea Finn, Niccolo Fusai, Manuel Y. Galliker, Dibya Ghosh, Lachy Groom, Karol Hausman, Brian Ichter, Szymon Jakubczak, Tim Jones, Liyiming Ke, Devin LeBlanc, Sergey Le... on Pith
- [5] Kevin Black, Noah Brown, Danny Driess, Adnan Esmail, Michael Robert Equi, Chelsea Finn, Niccolo Fusai, Lachy Groom, Karol Hausman, Brian Ichter, Szymon Jakubczak, Tim Jones, Liyiming Ke, Sergey Levine, Adrian Li-Bell, Mohith Mothukuri, Suraj Nair, Karl Pertsch, Lucy Xiaoyang S... on Pith
- [6] Anthony Brohan, Noah Brown, Justice Carbajal, Yevgen Chebotar, Xi Chen, Krzysztof Choromanski, Tianli Ding, Danny Driess, Avinava Dubey, Chelsea Finn, Pete Florence, Chuyuan Fu, Montse Gonzalez Arenas, Keerthana Gopalakrishnan, Kehang Han, Karol Hausman, Alexander Herzog, Jasm... on Pith
- [7] Siyi Chen, Hugo Hadfield, Alex Zook, Mikaela Angelina Uy, Chan Hee Song, Erwin Coumans, Xuning Yang, Faisal Ladhak, Qing Qu, Stan Birchfield, Jonathan Tremblay, and Valts Blukis. VoLo : A physical orchestrator for open-vocabulary long-horizon manipulation. arXiv preprint arXiv... on PithPith's reviewarXiv
- [8] Dongzhou Cheng, Taoran Yi, Ye Fang, Xingwu Zhang, Fan Feng, Yixuan Li, Gengxiong Zhuang, Rongze Wang, Shuai Yang, Wei Song, Weizhi Xue, Minyan Wu, Jie Gui, Jiaqi Wang, and Tong Wu. In-context robot learning with VLM agents. arXiv preprint arXiv:2609.19138, 2026 on PithPith's reviewarXiv
- [9] Max Fu, Justin Yu, Karim El-Refai, Ethan Kou, Haoru Xue, Huang Huang, Wenli Xiao, Guanzhi Wang, Fei-Fei Li, Guanya Shi, Jiajun Wu, Shankar Sastry, Yuke Zhu, Ken Goldberg, and Linxi Jim Fan. CaP-X : A framework for benchmarking and improving coding agents for robot manipulation... on Pith
- [10] Wenlong Huang, Fei Xia, Ted Xiao, Harris Chan, Jacky Liang, Pete Florence, Andy Zeng, Jonathan Tompson, Igor Mordatch, Yevgen Chebotar, Pierre Sermanet, Noah Brown, Tomas Jackson, Linda Luu, Sergey Levine, Karol Hausman, and Brian Ichter. Inner monologue: Embodied reasoning th... on Pith
- [11] Yash Jangir, Yidi Zhang, Kashu Yamazaki, Chenyu Zhang, Kuan-Hsun Tu, Tsung-Wei Ke, Lei Ke, Yonatan Bisk, and Katerina Fragkiadaki. RobotArena : Scalable robot benchmarking via real-to-sim translation. In International Conference on Learning Representations (ICLR), 2026 on Pith
- [12] Alexander Khazatsky, Karl Pertsch, Suraj Nair, Ashwin Balakrishna, Sudeep Dasari, Siddharth Karamcheti, Soroush Nasiriany, Mohan Kumar Srirama, Lawrence Yunliang Chen, Kirsty Ellis, Peter David Fagan, Joey Hejna, Masha Itkina, Marion Lepert, Yecheng Jason Ma, Patrick Tree Mill... on Pith
- [13] Xuanlin Li, Kyle Hsu, Jiayuan Gu, Karl Pertsch, Oier Mees, Homer Rich Walke, Chuyuan Fu, Ishikaa Lunawat, Isabel Sieh, Sean Kirmani, Sergey Levine, Jiajun Wu, Chelsea Finn, Hao Su, Quan Vuong, and Ted Xiao. Evaluating real-world robot manipulation policies in simulation. In Co... on Pith
- [14] Yi Li, Yuquan Deng, Jesse Zhang, Joel Jang, Marius Memmel, Raymond Yu, Caelan Garrett, Fabio Ramos, Dieter Fox, Anqi Li, Abhishek Gupta, and Ankit Goyal. HAMSTER : Hierarchical action models for open-world robot manipulation. In International Conference on Learning Representat... on Pith
- [15] Jacky Liang, Wenlong Huang, Fei Xia, Peng Xu, Karol Hausman, Brian Ichter, Pete Florence, and Andy Zeng. Code as policies: Language model programs for embodied control. In IEEE International Conference on Robotics and Automation (ICRA), 2023 on Pith
- [16] Bo Liu, Yifeng Zhu, Chongkai Gao, Yihao Feng, Qiang Liu, Yuke Zhu, and Peter Stone. LIBERO : Benchmarking knowledge transfer for lifelong robot learning. In Advances in Neural Information Processing Systems (NeurIPS), Datasets and Benchmarks Track, 2023 on Pith
- [17] Oier Mees, Lukas Hermann, Erick Rosete-Beas, and Wolfram Burgard. CALVIN : A benchmark for language-conditioned policy learning for long-horizon robot manipulation tasks. IEEE Robotics and Automation Letters (RA-L), 2022 on Pith
- [18] NVIDIA . Cosmos 3 : Omnimodal world models for physical AI . arXiv preprint arXiv:2606.02800, 2026 on PithPith's reviewarXiv
- [19] William Shen, Nishanth Kumar, Sahit Chintalapudi, Ryan Lindeborg, Jie Wang, Christopher Watson, Edward Hu, Jing Cao, Dinesh Jayaraman, Leslie Pack Kaelbling, and Tom \'a s Lozano-P \'e rez. TiPToP : A modular open-vocabulary robot manipulation system that plans. arXiv preprint... on PithPith's reviewarXiv
- [20] Lucy Xiaoyang Shi, Brian Ichter, Michael Robert Equi, Liyiming Ke, Karl Pertsch, Quan Vuong, James Tanner, Anna Walling, Haohuan Wang, Niccolo Fusai, Adrian Li-Bell, Danny Driess, Lachy Groom, Sergey Levine, and Chelsea Finn. Hi robot: Open-ended instruction following with hie... on Pith
show 3 more
- [21] Homer Walke, Kevin Black, Abraham Lee, Moo Jin Kim, Max Du, Chongyi Zheng, Tony Zhao, Philippe Hansen-Estruch, Quan Vuong, Andre He, Vivek Myers, Kuan Fang, Chelsea Finn, and Sergey Levine. BridgeData V2 : A dataset for robot learning at scale. In Conference on Robot Learning ... on Pith
- [22] Ruicheng Wang, Sicheng Xu, Yue Dong, Yu Deng, Jianfeng Xiang, Zelong Lv, Guangzhong Sun, Xin Tong, and Jiaolong Yang. MoGe-2 : Accurate monocular geometry with metric scale and sharp details. In Advances in Neural Information Processing Systems (NeurIPS), 2025 on Pith
- [23] Xuning Yang, Rishit Dagli, Alex Zook, Hugo Hadfield, Ankit Goyal, Stan Birchfield, Fabio Ramos, and Jonathan Tremblay. RoboLab : A high-fidelity simulation benchmark for analysis of task generalist policies. In Robotics: Science and Systems (RSS), 2026 on Pith
Cited by 1 Pith paper
Reviewed papers on Pith that cite this one. The full citation record
Challenges and discussion
Disagree with the review, or know something it missed? Leave a comment, or challenge this review formally: a challenge becomes a permanent entry in this paper's record, and its decision is published here.
Come back by email
A short list of the week's newly reviewed papers. Nothing is sent until you confirm from your inbox, and every email has a link that stops it. No account.
No comments yet.
Sign in to comment