aifollow.news
← TrendingLatest development Nov 06 16:00
STORY IN FOCUS

ToyBlocks releases robot-learning simulation framework and reports physical-robot tests

ToyBlocks brings task and demonstration generation, training, and evaluation into a controlled tabletop block simulation, with simulation assets, data, models, and code released. A Molmo 2 and ManiFlow system trained entirely on ToyBlocks data succeeded in 50.0% (15/30) of physical-robot episodes in the easiest three-block spelling setup; the researchers say more trials are needed to establish confidence across tasks.

1 developments2 reports2 sources

Event developments

Coverage of the same occurrence shares a node; subsequent developments have their own nodes.

Ai2 releases ToyBlocks, a simulation framework for robot learning

Ai2 has released ToyBlocks, a controlled tabletop block environment for generating training demonstrations and testing robot behavior. The initial dataset contains 85,336 long-horizon demonstrations. A system trained entirely on simulation data succeeded in 15 of 30 real-world episodes; the team says more trials are needed to establish statistical confidence.

View source reports · 2 sources · 2 reports
All reports · 2

By original publication time, with each report's bookmarks and feedback preserved.

Hugging Face 官方博客 Machine translation

ToyBlocks releases robot-learning simulation framework and reports physical-robot tests

ToyBlocks brings task and demonstration generation, training, and evaluation into a controlled tabletop block simulation, with simulation assets, data, models, and code released. A Molmo 2 and ManiFlow system trained entirely on ToyBlocks data succeeded in 50.0% (15/30) of physical-robot episodes in the easiest three-block spelling setup; the researchers say more trials are needed to establish confidence across tasks.

Allen Institute for AI 研究动态 Machine translation

Ai2 releases ToyBlocks, a simulation framework for robot learning

Ai2 has released ToyBlocks, a controlled tabletop block environment for generating training demonstrations and testing robot behavior. The initial dataset contains 85,336 long-horizon demonstrations. A system trained entirely on simulation data succeeded in 15 of 30 real-world episodes; the team says more trials are needed to establish statistical confidence.