Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
102 tokens/sec
GPT-4o
59 tokens/sec
Gemini 2.5 Pro Pro
43 tokens/sec
o3 Pro
6 tokens/sec
GPT-4.1 Pro
50 tokens/sec
DeepSeek R1 via Azure Pro
28 tokens/sec
2000 character limit reached

BridgeData V2: A Dataset for Robot Learning at Scale (2308.12952v3)

Published 24 Aug 2023 in cs.RO and cs.LG

Abstract: We introduce BridgeData V2, a large and diverse dataset of robotic manipulation behaviors designed to facilitate research on scalable robot learning. BridgeData V2 contains 60,096 trajectories collected across 24 environments on a publicly available low-cost robot. BridgeData V2 provides extensive task and environment variability, leading to skills that can generalize across environments, domains, and institutions, making the dataset a useful resource for a broad range of researchers. Additionally, the dataset is compatible with a wide variety of open-vocabulary, multi-task learning methods conditioned on goal images or natural language instructions. In our experiments, we train 6 state-of-the-art imitation learning and offline reinforcement learning methods on our dataset, and find that they succeed on a suite of tasks requiring varying amounts of generalization. We also demonstrate that the performance of these methods improves with more data and higher capacity models, and that training on a greater variety of skills leads to improved generalization. By publicly sharing BridgeData V2 and our pre-trained models, we aim to accelerate research in scalable robot learning methods. Project page at https://rail-berkeley.github.io/bridgedata

PDF HTML Abstract

An Overview of "BridgeData V2: A Dataset for Robot Learning at Scale"

"BridgeData V2: A Dataset for Robot Learning at Scale" introduces a comprehensive dataset designed to facilitate scalable robot learning research. As the capabilities of machine learning models improve, the need for extensive and diverse datasets becomes more apparent, particularly in fields like robotic manipulation that require models to generalize across various environments and tasks. This paper addresses a crucial aspect of such scalable learning: the provision of a dataset that significantly improves upon existing resources regarding size, diversity, and the potential for broad applicability.

Key Contributions and Methodological Insights

BridgeData V2 is notable for its scale and versatility, comprising 60,096 trajectories across 24 diverse environments. Compared to its predecessor, the original Bridge Dataset, BridgeData V2 includes more than seven times the number of demonstrations and spans a broader diversity of tasks and environments. The dataset supports a wide variety of skill sets, including fundamental manipulation tasks like pick-and-place and pushing, along with more intricate behaviors such as sweeping, folding, and environment alterations like opening drawers and doors.

A notable feature of the dataset is its compatibility with various open-vocabulary, multi-task learning methods, conditioned on goal images or natural language instructions. This compatibility is illustrated in the paper through the training of six state-of-the-art imitation learning and offline reinforcement learning methods, showcasing their success on tasks requiring different levels of generalization. The researchers demonstrate that performance improves with larger data quantities and model capacities, evidencing the utility of diverse data for enhancing generalization capabilities.

The paper also emphasizes the dataset's design considerations aimed at fostering reproducibility and usability across institutions, addressing a common barrier in robotic learning. By using a low-cost, publicly accessible robot platform, the paper ensures that researchers can train and exploit models without the need to recreate specific experimental setups meticulously.

Implications for Robotic Learning Research

The release of BridgeData V2 has significant implications for both the theoretical and practical aspects of robotic learning. Practically, this dataset serves as a valuable resource for developing and benchmarking new algorithms under realistic conditions, thanks to its extensive task and environment variability. Theoretically, BridgeData V2 is a step towards realizing scalable robot learning, emulating strategies seen in other AI fields, such as computer vision and natural language processing, which benefit from large, diverse datasets.

The dataset's impact is further amplified by its adaptability to different research methodologies. By evaluating a range of offline learning methods, the authors illustrate BridgeData V2’s adaptability, effectively supporting methods with diverse assumptions. This adaptability suggests the dataset's potential as a standard benchmarking tool within the community.

Future Directions and Considerations

While BridgeData V2 marks a considerable improvement over prior datasets, the paper acknowledges certain limitations that can inform future work. Tasks in the dataset, while varied, are primarily low-precision and do not encapsulate challenges faced in dynamic or high-force manipulation scenarios, such as in industrial environments. Additionally, although collected at a single institution with efforts to include diverse environments, future datasets could enhance validity by expanding environmental diversity further.

Future endeavors might also focus on multi-robot datasets that accommodate variability not only in environments and tasks but also across robot morphologies. This would enable a more robust evaluation of learning methods' transferability across different physical platforms.

In conclusion, BridgeData V2 significantly advances resources for scalable robot learning and offers a promising foundation for further exploration and innovation in the domain. Its scalability, diverse applicability, and open accessibility suggest it will be a pivotal dataset in the pursuit of more generalized and adaptable robotic systems.

File Document Download Save Streamline Icon: https://streamlinehq.com PDF File Document Download Save Streamline Icon: https://streamlinehq.com Markdown Bookmark Chat Bubble Oval Streamline Icon: https://streamlinehq.com Chat (Pro)
Definition Search Book Streamline Icon: https://streamlinehq.com
References (60)
  1. Language models are few-shot learners. Advances in neural information processing systems, 33:1877–1901, 2020.
  2. Zero-shot text-to-image generation. In International Conference on Machine Learning, pages 8821–8831. PMLR, 2021.
  3. Multiple interactions made easy (mime): Large scale demonstrations data for imitation. In Conference on robot learning, pages 906–915. PMLR, 2018.
  4. Roboturk: A crowdsourcing platform for robotic skill learning through imitation. In Conference on Robot Learning, pages 879–893. PMLR, 2018.
  5. BC-z: Zero-shot task generalization with robotic imitation learning. In 5th Annual Conference on Robot Learning, 2021. URL https://openreview.net/forum?id=8kbp23tSGYv.
  6. Bridge data: Boosting generalization of robotic skills with cross-domain datasets. arXiv preprint arXiv:2109.13396, 2021.
  7. Rt-1: Robotics transformer for real-world control at scale. arXiv preprint arXiv:2212.06817, 2022.
  8. L. P. Kaelbling. Learning to Achieve Goals. In IJCAI, 1993.
  9. Universal Value Function Approximators. In International Conference on Machine Learning (ICML), pages 1312–1320, 2015. ISBN 9781510810587. URL http://proceedings.mlr.press/v37/schaul15.pdf%****␣main.bbl␣Line␣75␣****http://jmlr.org/proceedings/papers/v37/schaul15.html.
  10. End-to-End Training of Deep Visuomotor Policies. Journal of Machine Learning Research (JMLR), 17(1):1334–1373, 2016. ISSN 15337928. doi:10.1007/s13398-014-0173-7.2. URL https://arxiv.org/pdf/1504.00702.pdf.
  11. J. Ho and S. Ermon. Generative Adversarial Imitation Learning. In Advances in Neural Information Processing Systems, 2016. URL https://arxiv.org/pdf/1606.03476.pdf.
  12. Visual foresight: Model-based deep reinforcement learning for vision-based robotic control. arXiv preprint arXiv:1812.00568, 2018.
  13. Deep imitation learning for complex manipulation tasks from virtual reality teleoperation. In 2018 IEEE International Conference on Robotics and Automation (ICRA), pages 5628–5635. IEEE, 2018.
  14. COG: Connecting New Skills to Past Experience with Offline Reinforcement Learning. In Conference on Robot Learning (CoRL), 2020.
  15. What matters in learning from offline human demonstrations for robot manipulation. In arXiv preprint arXiv:2108.03298, 2021.
  16. How to spend your robot time: Bridging kickstarting and offline reinforcement learning for vision-based robotic manipulation. 2022.
  17. Learning fine-grained bimanual manipulation with low-cost hardware. arXiv preprint arXiv:2304.13705, 2023.
  18. Diffusion policy: Visuomotor policy learning via action diffusion. arXiv preprint arXiv:2303.04137, 2023.
  19. L. Pinto and A. Gupta. Supersizing Self-supervision: Learning to Grasp from 50K Tries and 700 Robot Hours. IEEE International Conference on Robotics and Automation (ICRA), 2016.
  20. Learning Hand-Eye Coordination for Robotic Grasping with Deep Learning and Large-Scale Data Collection. International Journal of Robotics Research, 2017.
  21. Scalable Deep Reinforcement Learning for Vision-Based Robotic Manipulation. In Conference on Robot Learning, pages 651–673, 2018.
  22. Robot learning in homes: Improving generalization and reducing dataset bias. 7 2018. URL http://arxiv.org/abs/1807.07049.
  23. Robonet: Large-scale multi-robot learning. arXiv preprint arXiv:1910.11215, 2019.
  24. Learning to Poke by Poking: Experiential Learning of Intuitive Physics. In Advances in Neural Information Processing Systems, 2016. URL http://arxiv.org/abs/1606.07419.
  25. L. Pinto and A. Gupta. Learning to push by grasping: Using multiple tasks for effective learning. In 2017 IEEE international conference on robotics and automation (ICRA), pages 2161–2168. IEEE, 2017.
  26. Unsupervised learning for physical interaction through video prediction. Advances in neural information processing systems, 29, 2016.
  27. Combining self-supervised learning and imitation for vision-based rope manipulation. In 2017 IEEE international conference on robotics and automation (ICRA), pages 2146–2153. IEEE, 2017.
  28. One-shot imitation from observing humans via domain-adaptive meta-learning. arXiv preprint arXiv:1802.01557, 2018.
  29. Scaling Data-Driven Robotics With Reward Sketching and Batch Reinforcement Learning. arXiv preprint arXiv:1909.12200, 2019.
  30. MT-Opt: Continuous Multi-Task Robotic Reinforcement Learning at Scale. 2021.
  31. Actionable Models: Unsupervised Offline Reinforcement Learning of Robotic Skills. 4 2021.
  32. Roboagent: Towards sample efficient robot manipulation with semantic augmentations and action chunking. arxiv, 2023.
  33. Robocat: A self-improving foundation agent for robotic manipulation. arXiv preprint arXiv:2306.11706, 2023.
  34. Grasping in the wild: Learning 6dof closed-loop grasping from low-cost demonstrations. IEEE Robotics and Automation Letters, 5(3):4978–4985, 2020.
  35. Visual imitation made easy. In Conference on Robot Learning, pages 1992–2005. PMLR, 2021.
  36. Socially compliant navigation dataset (scand): A large-scale dataset of demonstrations for social navigation. IEEE Robotics and Automation Letters, 2022.
  37. Tartandrive: A large-scale dataset for learning off-road dynamics models. In 2022 International Conference on Robotics and Automation (ICRA), pages 2546–2552. IEEE, 2022.
  38. Gonet: A semi-supervised deep learning approach for traversability estimation. In 2018 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pages 3044–3051. IEEE, 2018.
  39. Gnm: A general navigation model to drive any robot. arXiv preprint arXiv:2210.03370, 2022.
  40. Meta-world: A benchmark and evaluation for multi-task and meta reinforcement learning. In Conference on robot learning, pages 1094–1100. PMLR, 2020.
  41. Maniskill: Generalizable manipulation skill benchmark with large-scale demonstrations. arXiv preprint arXiv:2107.14483, 2021.
  42. Rlbench: The robot learning benchmark & learning environment. IEEE Robotics and Automation Letters, 5(2):3019–3026, 2020.
  43. Habitat: A platform for embodied ai research. In Proceedings of the IEEE/CVF international conference on computer vision, pages 9339–9347, 2019.
  44. Ai2-thor: An interactive 3d environment for visual ai. arXiv preprint arXiv:1712.05474, 2017.
  45. Pre-training for robots: Offline rl enables learning new tasks from a handful of trials. arXiv preprint arXiv:2210.05178, 2022.
  46. Offline Reinforcement Learning With Implicit Q-Learning. arXiv preprint arXiv:2110.06169, 2021.
  47. Idql: Implicit q-learning as an actor-critic method with diffusion policies. arXiv preprint arXiv:2304.10573, 2023.
  48. Contrastive learning as goal-conditioned reinforcement learning. Advances in Neural Information Processing Systems, 35:35603–35620, 2022.
  49. Rh20t: A robotic dataset for learning diverse skills in one-shot. In RSS 2023 Workshop on Learning for Task and Motion Planning, 2023.
  50. Behavior transformers: Cloning k𝑘kitalic_k modes with one stone. In Thirty-Sixth Conference on Neural Information Processing Systems, 2022. URL https://openreview.net/forum?id=agTr-vRQsa.
  51. Denoising diffusion probabilistic models. Advances in Neural Information Processing Systems, 33:6840–6851, 2020.
  52. Learning Structured Output Representation using Deep Conditional Generative Models. In Advances in Neural Information Processing Systems, 2015.
  53. C-learning: Learning to achieve goals via recursive classification. In International Conference on Learning Representations, 2021. URL https://openreview.net/forum?id=tc5qisoB-C.
  54. Stabilizing contrastive rl: Techniques for offline goal reaching, 2023.
  55. Language-conditioned imitation learning for robot manipulation tasks. ArXiv, abs/2010.12083, 2020.
  56. Multilingual Universal Sentence Encoder for Semantic Retrieval, July 2019.
  57. FiLM: Visual Reasoning with a General Conditioning Layer, Dec. 2017.
  58. M. Tan and Q. Le. Efficientnet: Rethinking model scaling for convolutional neural networks. In International conference on machine learning, pages 6105–6114. PMLR, 2019.
  59. Universal sentence encoder for english. In Proceedings of the 2018 conference on empirical methods in natural language processing: system demonstrations, pages 169–174, 2018.
  60. D. Kingma and J. Ba. Adam: A method for stochastic optimization. International Conference on Learning Representations (ICLR), 2015.
User Edit Pencil Streamline Icon: https://streamlinehq.com
Authors (14)
  1. Homer Walke (14 papers)
  2. Kevin Black (29 papers)
  3. Abraham Lee (4 papers)
  4. Moo Jin Kim (9 papers)
  5. Max Du (1 paper)
  6. Chongyi Zheng (9 papers)
  7. Tony Zhao (3 papers)
  8. Philippe Hansen-Estruch (10 papers)
  9. Quan Vuong (41 papers)
  10. Andre He (11 papers)
  11. Vivek Myers (16 papers)
  12. Kuan Fang (30 papers)
  13. Chelsea Finn (264 papers)
  14. Sergey Levine (531 papers)
Citations (128)
View on Semantic Scholar
Github Logo Streamline Icon: https://streamlinehq.com

GitHub

  1. BridgeData V2
  2. GitHub - rail-berkeley/bridge_data_v2 (110 stars)