AlphaNIST
Researchers train one robot policy to complete all twenty NIST assembly-board tasks using success-guided sampling.
AlphaNIST describes training a single robot policy on NIST Assembly Task Board #1, which has twenty tasks and parts as small as 4 mm. The policy learns from a binary success reward and diverse simulated starts, including loose parts, in-hand grasps, and cached partial assemblies rather than human demonstrations. Success-guided beta sampling emphasizes states the policy solves about half the time. In a single-component experiment, one policy performs all twenty assembly tasks.
- One policy is trained across all twenty NIST assembly tasks.
- Parts are as small as 4 mm, with tight clearances that can jam.
- Training uses a binary success reward, not scripted demonstrations.
- Success-guided sampling favors states solved about half the time.
- Products
- NIST Assembly Task Board
- Organizations
- NIST
From the original article
1 min read at alpha-nist.github.io
The NIST board has twenty assembly tasks, with parts as small as 4 mm. Fitting one part can take several changes of grasp and orientation. Tight clearances leave little room for error: a tilted nut can jam or cross-thread.
By jasonjmcghee · alpha-nist.github.io
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.