Ai2 releases ToyBlocks, an open simulated tabletop for robot learning experiments
Article timeline
The articles in this story. Times are ET.
Ai2OfficialAi2 describes ToyBlocks' interface for connecting frontier models to robotsAIAi2 says ToyBlocks includes a tool interface that connects a frontier model, letting users instruct the robot in natural language. Through camera images and skill calls, researchers can study how the model searches, plans, and handles mistakes.
- Ai2OfficialAi2 introduces ToyBlocks, a simulated robot reasoning setting with blocks
AIAi2 introduces ToyBlocks, a simulated setting where a robot must reason about how blocks interact as it pushes, flips, and stacks them. The setting mirrors how young children learn spatial planning and reasoning through play with toy blocks.
- Ai2OfficialAi2 introduces ToyBlocks, an open simulated tabletop for robot learning
AIAi2 introduces ToyBlocks, an open simulated tabletop for controlled experiments in robot learning. The tool lets researchers control training data and models, run targeted evals, and trace failures to their causes.
- Hugging Face BlogOfficialAi2 introduces ToyBlocks, a controlled simulation for robot learning research
AIAi2 introduces ToyBlocks, an open simulation framework with a robot arm and blocks for studying robot learning and evaluation. It automatically generates tasks and demonstrations, and it reports 85,336 long-horizon demonstrations across nearly 1,600 hours of execution data and 144 million frames. The post also reports tests of VLA spatial generalization, high-level planning, and world models, with a real-robot comparison showing 50.0% success on 30 episodes versus 55.5% in simulation.
- Ai2 (Allen Institute for AI)OfficialPickAi2 introduces ToyBlocks, a controlled simulation for robot learning
AIAi2 introduces ToyBlocks, an open simulation framework built around a robot arm and blocks on a tabletop. It automatically generates demonstrations and metrics so researchers can vary training data and trace failures. Ai2 releases 85,336 demonstrations covering nearly 1,600 hours of robot execution data, along with training and evaluation code.