mlsanityData to 7 Oct 2026

When Dynamics Shift, Robust Task Inference Wins: Offline Imitation Learning with Behavior Foundation Models Revisited

arXiv 2605.17017 ↗First posted 16 May 2026cs.LG cs.AIcited by 0

On the map: Offline RL and imitation learning from data → Imitation learning methods

Abstract (arXiv)

Behavior Foundation Models (BFMs) enable scalable imitation learning (IL) by pretraining task-agnostic representations that can be rapidly adapted to new tasks. However, existing BFMs assume fixed environment dynamics, limiting their robustness under real-world shifts such as changes in friction, actuation, or sensor noise. We address this by formulating BFM task-inference as a robust minimax optimization problem, enabling adaptation to worst-case dynamics perturbations without modifying pretraining. To the best of our knowledge, this is the first BFM-based framework that achieves robustness to dynamics shifts while relying solely on offline data from a single nominal environment. Our approach significantly outperforms standard BFM and robust offline IL baselines under dynamics shifts. These results demonstrate that robust policy can be achieved entirely at task-inference time, improving the practicality of BFMs in dynamic settings.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Placed in Imitation learning methods, one of 13 directions in Offline RL and imitation learning from data. Builds on 46 of its 63 references that are in the map. First posted on arXiv 16 May 2026.

Builds on

46 of 63 references.

Closest ancestors · 9

Same direction, newest first

  1. 2025Balance Equation-based Distributionally Robust Offline Imitation Learning11 Nov
  2. 2025Robust Behavior Cloning Via Global Lipschitz Regularization24 Jun
  3. 2024Markov Balance Satisfaction Improves Performance in Strictly Batch Offline Imitation Learning17 Aug
  4. 2023Conditional Kernel Imitation Learning for Continuous State Environments24 Aug
  5. 2022Robust Imitation Learning against Variations in Environment Dynamics19 Jun
  6. 2020Robust Imitation Learning from Noisy Demonstrations20 Oct
  7. 2019Imitation Learning via Off-Policy Distribution Matching10 Dec
  8. 2019Watch, Try, Learn: Meta-Learning from Demonstrations and Reward7 Jun
  9. 2019Imitation Learning from Imperfect Demonstration27 Jan

Other references · 37

  1. 2023GPT-4 Technical Reportcited by 18,132
  2. 2023Direct Preference Optimization: Your Language Model is Secretly a Reward Modelcited by 7,916
  3. 2016Layer Normalizationcited by 5,579
  4. 2021Convex optimizationcited by 2,479
  5. 2016Generative Adversarial Imitation Learningcited by 1,576
  6. 2018Off-Policy Deep Reinforcement Learning without Explorationcited by 1,227
  7. 2018Exploration by Random Network Distillationcited by 966
  8. 2018DeepMind Control Suitecited by 935
  9. 2017Sim-to-Real Transfer of Robotic Control with Dynamics Randomizationcited by 616
  10. 2019Exploring the Limitations of Behavior Cloning for Autonomous Drivingcited by 343
  11. 2016Successor Features for Transfer in Reinforcement Learningcited by 331
  12. 2017One-Shot Imitation Learningcited by 232
25 more
  1. 2017One-Shot Visual Imitation Learning via Meta-Learningcited by 225
  2. 2017DART: Noise Injection for Robust Imitation Learningcited by 187
  3. 2021Reinforcement Learning with Prototypical Representationscited by 179
  4. 2021Learning One Representation to Optimize All Rewardscited by 110
  5. 2021APS: Active Pretraining with Successor Featurescited by 109
  6. 2018Universal Successor Features Approximatorscited by 102
  7. 2022Robust Reinforcement Learning using Offline Datacited by 95
  8. 2022Don't Change the Algorithm, Change the Data: Exploratory Data for Offline Reinforcement Learningcited by 93
  9. 2024Foundation Policies with Hilbert Representationscited by 89
  10. 2022Does Zero-Shot Reinforcement Learning Exist?cited by 79
  11. 2021Learning Successor States and Goal-Dependent Values: A Mathematical Viewpointcited by 67
  12. 2018Task-Embedded Control Networks for Few-Shot Imitation Learningcited by 65
  13. 2021Generalization in Dexterous Manipulation via Geometry-Aware Multi-Task Learningcited by 49
  14. 2020Distributional Robustness and Regularization in Reinforcement Learningcited by 39
  15. 2025The Reality Gap in Robotics: Challenges, Solutions, and Best Practicescited by 22
  16. 2023Zero-Shot Reinforcement Learning from Low Quality Datacited by 21
  17. 2024ODICE: Revealing the Mystery of Distribution Correction Estimation via Orthogonal-gradient Updatecited by 21
  18. 2025Dexterous Manipulation through Imitation Learning: A Surveycited by 18
  19. 2023Bridging Distributionally Robust Learning and Offline RL: An Approach to Mitigate Distribution Shift and Partial Data Coveragecited by 12
  20. 2025Zero-Shot Adaptation of Behavioral Foundation Models to Unseen Dynamicscited by 8
  21. 2025Towards Robust Zero-Shot Reinforcement Learningcited by 7
  22. 2025Zero-Shot Reinforcement Learning Under Partial Observabilitycited by 6
  23. 2025Optimistic Task Inference for Behavior Foundation Modelscited by 5
  24. 2024Meta-Controller: Few-Shot Imitation of Unseen Embodiments and Tasks in Continuous Controlcited by 3
  25. 2025Impact of Static Friction on Sim2Real in Robotic Reinforcement Learningcited by 2

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2025Task Tokens: A Flexible Approach to Adapting Behavior Foundation Models
  2. 2021Mitigating Covariate Shift in Imitation Learning via Offline Data Without Great Coverage
  3. 2025Robust Offline Imitation Learning Through State-level Trajectory Stitching
  4. 2025RISE: Robust Imitation through Stochastic Encoding
  5. 2026It's Not Just More Demos: Counterfactual Action Sensitivity Coverage for Data-Efficient Robust Robot Imitation
  6. 2026On the Sample Efficiency of Inverse Dynamics Models for Semi-Supervised Imitation Learning
  7. 2024Learning on One Mode: Addressing Multi-modality in Offline Reinforcement Learning
  8. 2025Adapt Your Body: Mitigating Proprioception Shifts in Imitation Learning