# ReSkill boosts robot task success from 30% to over 80% in simulation

> The new selective self‑improvement framework raises average success on ten RoboDojo tasks to 86.2% and reaches 77.5% on real‑world trials.

Oossa · 2026-11-10 · https://oossa.com/en/reskill-boosts-robot-task-success-from-30-to-over-80-in-simulation

Event date: 2026-11-07

Researchers led by Zheng Sun introduced ReSkill, a system that lets a robot identify and fix weak spots in its manipulation programs. The framework builds a task graph, spots failing steps, and proposes small code changes that are tested before being kept. In simulated RoboDojo tasks the average success climbed from 30% to 86.2%, and on eight real‑world tasks the robot succeeded in 62 of 80 trials. The final policies run without the self‑improvement agent, using only the refined skills.

## The facts

- Average success on ten RoboDojo tasks rose from 30.0% to 86.2%
- Paper submitted on 2026-11-07

## Why it matters

Manufacturers can deploy robots that improve their own code during development, leading to more reliable pick‑and‑place operations without extra runtime overhead.

## Sources & references

1. [ReSkill: Selective Self-Improvement for Robot Manipulation](https://arxiv.org/abs/2611.10510) – arXiv cs.RO, 2026-11-10

Last updated: 2026-11-10