Agenticroborl boosts robot planners by post-training across different harnesses
AgenticRoboRL: Bringing Coding-Agent Post-Training to Agentic Robotics
Robotics
Summary
Robots often use special planners to decide how to perform complicated tasks, but these planners don’t always improve themselves. The authors introduce AgenticRoboRL, a way to train and improve these robot planners after they are initially built, by using a mix of guided learning and trial-and-error feedback. This method works even when different robot control systems are involved and showed a large increase in task success in tests. Their approach narrows the gap between smaller trained planners and much larger models that require extra memory.
What this means in practice
- •For industrial robot programmers: Improve robot task success by post-training planners to work seamlessly with different robot control systems.
- •For software engineers for automation: Enhance automation software by integrating AgenticRoboRL to increase robot planner effectiveness without changing execution hardware.
Authors
Xinxin Liu, Bingxin Xu, Wenrui Bao, Ming Li, Wentao Bao, Chen Chen, Yuzhang Shang
Abstract
Agentic robotics systems combine planners with robot harnesses to perform complex manipulation tasks. Recent work improves these systems through better harnesses and external memory, but relies on powerful closed-source models as planners and does not improve the planners' own capabilities. In contrast, coding agents have improved substantially by post-training the model itself. We therefore investigate whether post-training can also improve robot planners. However, applying it across robot harnesses requires accommodating different execution mechanisms, including VLA skill invocation and control-code generation. To this end, we introduce AgenticRoboRL, a framework for post-training robot planners across different harnesses. The training recipe combines supervised fine-tuning with multi-turn reinforcement learning. To accommodate different execution mechanisms, harness-specific adapters connect recorded planner interactions with task outcomes, enabling a common RL training pipeline while keeping execution components fixed. On LIBERO-PRO, AgenticRoboRL raises the success rate of a 9B-parameter planner from 18.1% to 71.0% without external memory, approaching the 72.1% achieved by Codex (GPT-5.5) with external memory. These results demonstrate post-training as an effective approach for improving robot planners. We will release the framework and trained planner checkpoints.