Robotic Programmer: Video Instructed Policy Code Generation for Robotic Manipulation
This addresses the problem of enabling robots to perform diverse manipulation tasks without task-specific training, though it builds incrementally on existing policy code generation methods.
The authors tackled zero-shot generalization in robotic manipulation by proposing Robotic Programmer (RoboPro), a foundation model that generates policy code from visual inputs and instructions, achieving a state-of-the-art zero-shot success rate on RLBench that surpasses GPT-4o by 11.6%.
Zero-shot generalization across various robots, tasks and environments remains a significant challenge in robotic manipulation. Policy code generation methods use executable code to connect high-level task descriptions and low-level action sequences, leveraging the generalization capabilities of large language models and atomic skill libraries. In this work, we propose Robotic Programmer (RoboPro), a robotic foundation model, enabling the capability of perceiving visual information and following free-form instructions to perform robotic manipulation with policy code in a zero-shot manner. To address low efficiency and high cost in collecting runtime code data for robotic tasks, we devise Video2Code to synthesize executable code from extensive videos in-the-wild with off-the-shelf vision-language model and code-domain large language model. Extensive experiments show that RoboPro achieves the state-of-the-art zero-shot performance on robotic manipulation in both simulators and real-world environments. Specifically, the zero-shot success rate of RoboPro on RLBench surpasses the state-of-the-art model GPT-4o by 11.6%, which is even comparable to a strong supervised training baseline. Furthermore, RoboPro is robust to variations on API formats and skill sets.