ClaraChengGo(@ClaraChengGo):This is what robotics policy training is heading to. How it works now: engineer - reward function - mass training - success But Nvidia ASPIRE is more like: Receive a new task - Execute the policy - Evaluate the outcome - Identify failure patterns - Optimize the policy - Repeat - Master the task Self-improvement for robotics model.

ClaraChengGo

@ClaraChengGo

Babysitting robots with model-ready data at @openmind_agi @fabricFND

加入 June 2022

4.4K 正在关注 6.8K 粉丝

ClaraChengGo@ClaraChengGo

2026.06.30 19:02

This is what robotics policy training is heading to. How it works now: engineer - reward function - mass training - success But Nvidia ASPIRE is more like: Receive a new task - Execute the policy - Evaluate the outcome - Identify failure patterns - Optimize the policy - Repeat - Master the task Self-improvement for robotics model.

显示更多