This research introduces Robotic Policy Inference via Synthetic Observations (ROSO), a novel method to improve the performance of pre-trained policies in robotic systems. ROSO uses a technique called Stable Diffusion to pre-process a robot’s observations of new objects to fit within its distribution of observations of the pre-trained policies. This method allows the transfer of learned knowledge from known tasks to unseen scenarios, enhancing the robot’s adaptability without requiring extensive fine-tuning. The study shows that incorporating generative AI into robotic inference significantly improves successful outcomes, finishing up to 57% of tasks otherwise unsuccessful with the pre-trained policy.
Publication date: 28 Nov 2023
Project Page: https://yusuke710.github.io/roso.github.io/
Paper: https://arxiv.org/pdf/2311.16680
