Energy-Regularized Imitation Learning for Force- and Work-Aware Robotic Manipulation
On the map: Offline RL and imitation learning from data → Robot manipulation imitation learning
Abstract (arXiv)
This paper studies energy-aware manipulation as a physically grounded learning problem. We define a joint-space mechanical-work proxy from joint torque and angular displacement, and train a differentiable energy predictor that estimates this work from robot states and actions. The predictor converts a non-differentiable simulator-side physical quantity into a differentiable regularizer for fine-tuning a pretrained manipulation policy. We instantiate the framework with RVT-2 on RLBench and evaluate 12 manipulation tasks involving object contact, articulated motion, placement, pushing, and sweeping. The proposed fine-tuning reduces the average mechanical work from 208.8J to 204.4J (i.e., 2.1% reduction), while the mean task success rate also increases slightly from 86.2% to 86.9%. These results show that work-aware policy optimization can suppress physically inefficient motion without requiring an explicit differentiable dynamics model.
Key facts. No citing papers in the map yet (data through 7 Oct 2026). Placed in Robot manipulation imitation learning, one of 13 directions in Offline RL and imitation learning from data. Builds on 17 of its 25 references that are in the map. First posted on arXiv 16 Sep 2026.
Builds on
17 of 25 references.
References · 17
- 2019Meta-World: A Benchmark and Evaluation for Multi-Task and Meta Reinforcement Learningcited by 1,168
- 2019Large Batch Optimization for Deep Learning: Training BERT in 76 minutescited by 671
- 2019RLBench: The Robot Learning Benchmark & Learning Environmentcited by 669
- 2022Perceiver-Actor: A Multi-Task Transformer for Robotic Manipulationcited by 545
- 2017Deep learning for universal linear embeddings of nonlinear dynamicscited by 491
- 2017Snapshot Ensembles: Train 1, get M for freecited by 349
- 2019DiffTaichi: Differentiable Programming for Physical Simulationcited by 248
- 2024The Road Less Scheduledcited by 158
- 2024RVT-2: Learning Precise Manipulation from Few Demonstrationscited by 135
- 2023Q-Transformer: Scalable Offline Reinforcement Learning via Autoregressive Q-Functionscited by 129
- 2022Deep Whole-Body Control: Learning a Unified Policy for Manipulation and Locomotioncited by 98
- 2022Deep Black-Box Reinforcement Learning with Movement Primitivescited by 18
5 more
- 2023Data-Driven Stochastic Motion Evaluation and Optimization with Image by Spatially-Aligned Temporal Encodingcited by 3
- 2023R2-Diff: Denoising by diffusion as a refinement of retrieved motion for image-based motion predictioncited by 3
- 2025Robot Motion Planning using One-Step Diffusion with Noise-Optimized Approximate Motionscited by 3
- 2026Energy-Aware Reinforcement Learning for Robotic Manipulation of Articulated Components in Infrastructure Operation and Maintenancecited by 3
- 2026Energy-Efficient Arm Reaching for a Humanoid Robot via Deep Reinforcement Learning with Identified Power Modelscited by 1
Built on by
Not cited yet.
Similar, not cited
Close in wording, no citation either way
- 2026Contact-Aware Imitation Learning Through Contact Factorization
- 2025FLEX: A Framework for Learning Robot-Agnostic Force-based Skills Involving Sustained Contact Object Manipulation
- 2026Wrench-ACT: Enhancing Robot Policies for Contact Rich Behavior Using Direct Wrench Control
- 2025Bootstrapping Imitation Learning for Long-horizon Manipulation via Hierarchical Data Collection Space
- 2025Dexterous Manipulation through Imitation Learning: A Survey
- 2026One-Shot Learning from Demonstration of Contact-Rich Robotic Manipulation by Identifying Physical Interactions
- 2025DexForce: Extracting Force-informed Actions from Kinesthetic Demonstrations for Dexterous Manipulation
- 2026A Hierarchical Approach to Imitation Learning for Manipulation Tasks Requiring Time Varying Forces