DenseReward: Dense Reward Learning via Failure Synthesis for Robotic Manipulation
DenseReward: Dense Reward Learning via Failure Synthesis for Robotic Manipulation
This research presents DenseReward, a dense robotic reward model designed to overcome limitations in reinforcement learning by synthesizing physically realistic failure trajectories without manual annotation. The system predicts frame-level reward scores from visual observations and language instructions, enabling fine-grained task progress estimation. Experiments show DenseReward outperforms general-purpose vision-language models and existing robotic reward models in both simulated and real-world manipulation tasks, providing effective guidance for downstream model predictive control and reinforcement learning.
The paper introduces DenseReward, a vision-language reward model trained on synthesized failure trajectories to provide dense frame-level feedback for robotic manipulation, demonstrating improved performance over general-purpose VLMs and existing models in both simulation and real-world settings.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
- paper: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
Source：arXiv Robotics — research abstracts · arxiv.org