RoboWits - Mutations: leaderboard

RoboWits (Lin et al., 2026) is a simulated bimanual manipulation benchmark of creative problem solving: seed tasks need geometry, material or assembly reasoning, such as using a rod to reach an object or displacing water to raise its level. This board scores the generated mutations of 10 seed tasks, each adding a constraint that breaks the usual solution, for policies trained only on the seed tasks' human demonstrations; success rate in percent over 5 trials per mutation, higher is better.

Metric: Task success rate (%; mean over the mutations of the 10 evaluated seed tasks, each mutation adding a constraint that needs a new solution; 5 randomly placed trials per mutation; policy trained or fine-tuned on the 50 human teleoperation demonstrations of each of the 10 seed tasks, none of a mutation). Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 3 models tracked.

Top models

Interactive version: theaggregate.ai/benchmark?slug=robowits-mutations · How It Works · Data refreshed daily, snapshot 2026-11-12.