AI2 Releases ToyBlocks, a Simulator for Training Robots
AI2 has released ToyBlocks, a simulated-world tool where robots practice picking up and stacking objects. Its ManiFlow model hit an 80-81% success rate.
The Allen Institute for Artificial Intelligence (AI2), a non-profit AI research group, has announced the release of their ToyBlocks platform for creating controlled environments. Using this environment, a robotic arm can learn how to manipulate blocks in simulation rather than in the real world. They’ve also made available 85,336 recorded demonstrations from this simulation, along with a model they’ve developed, ManiFlow, which was able to succeed in 80% of its trials in simulation and 81% of its trials on a real robot arm.
Learning in the real world can be expensive, time consuming, and difficult to reproduce across labs. Having a common simulation environment allows researchers to compare techniques and understand what went wrong during failure. This isn’t intended as a product for businesses looking for robots to deploy in warehouses or homes, but instead as a tool for other researchers developing robot control policies.
Source: huggingface.co
Free course
Stop reading about AI — start building with it
The free Claude Code course: your first site, tool or game — no coding. No upsells, no cross-sells — nothing to buy here.
Start free →
Author
Evgenii Arsentev
PhD · Founder & CEO, Dad4Kids
Articles · Latest articles