Generalizable Dense Reward for Long-Horizon Robotic Tasks
This addresses the challenge of generalizable reward design for long-horizon robotic tasks, offering a method to improve success rates and efficiency without manual engineering, though it is incremental as it builds on existing RL and foundation model techniques.
The paper tackles the problem of robotic foundation policies struggling with long-horizon tasks due to distribution shift and error accumulation, proposing VLLR, a dense reward framework that combines LLM/VLM-based extrinsic rewards and policy self-certainty intrinsic rewards, achieving up to 56% absolute success rate gains over pretrained policies and up to 5-10% gains over state-of-the-art RL methods on in-distribution and out-of-distribution tasks without manual reward engineering.
Existing robotic foundation policies are trained primarily via large-scale imitation learning. While such models demonstrate strong capabilities, they often struggle with long-horizon tasks due to distribution shift and error accumulation. While reinforcement learning (RL) can finetune these models, it cannot work well across diverse tasks without manual reward engineering. We propose VLLR, a dense reward framework combining (1) an extrinsic reward from Large Language Models (LLMs) and Vision-Language Models (VLMs) for task progress recognition, and (2) an intrinsic reward based on policy self-certainty. VLLR uses LLMs to decompose tasks into verifiable subtasks and then VLMs to estimate progress to initialize the value function for a brief warm-up phase, avoiding prohibitive inference cost during full training; and self-certainty provides per-step intrinsic guidance throughout PPO finetuning. Ablation studies reveal complementary benefits: VLM-based value initialization primarily improves task completion efficiency, while self-certainty primarily enhances success rates, particularly on out-of-distribution tasks. On the CHORES benchmark covering mobile manipulation and navigation, VLLR achieves up to 56% absolute success rate gains over the pretrained policy, up to 5% gains over state-of-the-art RL finetuning methods on in-distribution tasks, and up to $10\%$ gains on out-of-distribution tasks, all without manual reward engineering. Additional visualizations can be found in https://silongyong.github.io/vllr_project_page/