Hierarchical robot-control and evaluation platform
A robotics platform can divide long-horizon work between a subtask-planning policy and a generalist controller, using a world model to search alternative actions when plans become uncertain.
From Github Awesome — GitHub Trending Today #42: open-science, AgentENV, GodotHub, PGSimCity, rescript, agentacct, Crisp at 12:12
Problem: Long-horizon robot tasks can become unreliable when a single controller continues blindly after uncertainty appears.
For: Robotics researchers and teams training or evaluating general-purpose robot policies.
Examples
- tau-0-vla splits robot work into planning and execution levels, adds test-time alternative-action search through a world model, and includes post-training recipes, robot adapters, and open-loop evaluation.
Soon you can unlock the full business plan.
Behind this: 1 more real example · 3 things the video never answers.