HiRE: Hindsight Reward Editing for Policy Finetuning
cs.RO
Submitted: 2026-09-22
Updated: 2026-09-22
Comments: CoRL 2026
License: http://creativecommons.org/licenses/by-nc-nd/4.0/
The gist: Pre-trained robot policies always require finetuning to adapt to specific environments.
Terminology
Abstract
Pre-trained robot policies always require finetuning to adapt to specific environments. Reinforcement Learning (RL) offers high performance potential because it improves action optimality rather than simply mimicking data. However, such potential depends heavily on reward quality. Sparse rewards lack process feedback, human-designed rewards are costly and biased, and semantic rewards from foundation representations are often not control-centric. We propose Hindsight Reward Editing (HiRE), a training-free framework to break this reward bottleneck. HiRE bridges the broad knowledge of foundation representation models with physical control awareness, by contrasting successful and failed trajectories in hindsight. It calibrates foundation representation models by identifying "trap states" that are predicted as high-rewarding states yet eventually result in failure, and vice versa. HiRE explicitly penalizes these traps while boosting rewards for critical successful states. This approach can be flexibly compatible with any foundation representations and RL algorithms. Experiments show that HiRE consistently outperforms other reward recipes by delivering dense, control-aware feedback that prevents value function collapse and reward hacking, thereby achieving superior sample efficiency, stable policy updates, and higher performance ceilings, e.g., at least 3x performance of the base policies. Qualitative results are at https://hire-project.github.io.
Sources
- $\pi^{*}_{0.6}$: a VLA That Learns From Experience
- RL Token: Bootstrapping Online RL with Vision-Language-Action Models
- RoboReward: General-Purpose Vision-Language Reward Models for Robotics
- TOPReward: Token Probabilities as Hidden Zero-Shot Rewards for Robotics
- Robo-Dopamine: General Process Reward Modeling for High-Precision Robotic Manipulation
- VIP: Towards Universal Visual Reward and Representation via Value-Implicit Pre-Training
- SigLIP 2: Multilingual Vision-Language Encoders with Improved Semantic Understanding, Localization, and Dense Features
- From Prior to Pro: Efficient Skill Mastery via Distribution Contractive RL Finetuning
- Rewarding DINO: Predicting Dense Rewards with Vision Foundation Models
- $\pi_{0.5}$: a Vision-Language-Action Model with Open-World Generalization
- $\pi_\texttt{RL}$: Online RL Fine-tuning for Flow-based Vision-Language-Action Models
- Proximal Policy Optimization Algorithms
- Residual Off-Policy RL for Finetuning Behavior Cloning Policies
- robosuite: A Modular Simulation Framework and Benchmark for Robot Learning
- Efficient Online Reinforcement Learning with Offline Data
Related papers
- FMT x: An Efficient and Asymptotically Optimal Extension of the Fast Marching Tree for Dynamic Replanning
- MPCFormer: A physics-informed data-driven approach for explainable socially-aware autonomous driving
- RoboLab: A High-Fidelity Simulation Benchmark for Analysis of Task Generalist Policies
- ARC: A Reasoning Recipe for Robot Foundation Models
- HRDexDB: A 4D Dexterous Grasping Dataset Across Human and Multiple Robot Embodiments
- APT: Action Expert Pretraining Improves Instruction Generalization of Vision-Language-Action Policies