RoboRSI
A robot self-improvement system based on top-down skill refinement. Through a hierarchical skill structure, it attributes execution experience to specific capability branches, enabling stable reuse and iterative evolution of skills.
Timeline (last 90 days)
The research team proposes RoboRSI, a robot self-improvement system based on top-down skill refinement (TSR).
RoboRSI completes planning, execution, diagnosis, and the validation and release of new skills through the collaborative work of four roles: Manager, Planner, Engineer, and Reviewer.
On a mobile manipulation robot platform, RoboRSI developed multi-object household cleanup capability through 104 rounds of iterative development.
In simulation environments, RoboRSI achieves the highest success rates on benchmarks such as LIBERO, LIBERO-PRO, LIBERO-Plus, and RoboTwin, exceeding the strongest baseline by 2.7 to 11.0 percentage points.
RoboRSI's local revision and global stabilization mechanism both avoids catastrophic forgetting and makes the repair process traceable and verifiable, providing structural guarantees for the reuse and composition of subsequent skills.
Humans guide RoboRSI's evolution process through goals and corrections, ensuring that the system evolves in the right direction.
RoboRSI's way of organizing skills is not only applicable to a single scenario but can also transfer across tasks and environments, which is important evidence for evaluating whether a self-improvement system is truly general.
In the future, the research team may further explore how to extend RoboRSI to more types of tasks and environments, and how to reduce dependence on human guidance to achieve a higher degree of autonomous evolution.
All Facts (8)
The research team proposes RoboRSI, a robot self-improvement system based on top-down skill refinement (TSR).
50%UnverifiedRoboRSI completes planning, execution, diagnosis, and the validation and release of new skills through the collaborative work of four roles: Manager, Planner, Engineer, and Reviewer.
50%UnverifiedHumans guide RoboRSI's evolution process through goals and corrections, ensuring that the system evolves in the right direction.
50%UnverifiedRoboRSI's way of organizing skills is not only applicable to a single scenario but can also transfer across tasks and environments, which is important evidence for evaluating whether a self-improvement system is truly general.
50%UnverifiedIn simulation environments, RoboRSI achieves the highest success rates on benchmarks such as LIBERO, LIBERO-PRO, LIBERO-Plus, and RoboTwin, exceeding the strongest baseline by 2.7 to 11.0 percentage points.
50%UnverifiedRoboRSI's local revision and global stabilization mechanism both avoids catastrophic forgetting and makes the repair process traceable and verifiable, providing structural guarantees for the reuse and composition of subsequent skills.
50%UnverifiedOn a mobile manipulation robot platform, RoboRSI developed multi-object household cleanup capability through 104 rounds of iterative development.
50%UnverifiedIn the future, the research team may further explore how to extend RoboRSI to more types of tasks and environments, and how to reduce dependence on human guidance to achieve a higher degree of autonomous evolution.
50%