Unverified50% confidenceFactExact time
OpenAI在2019年训练Dactyl机械手时，在MuJoCo仿真中成功率超过90%的抓取策略在真实Shadow Hand上直接部署时成功率骤降至30%以下
1
Sources
50%
Confidence
Long-term
Relevance
7/17/2026
First Seen
Sources
Related Claims
UnverifiedIn 2019, OpenAI used domain randomization techniques to train a robotic hand to solve a Rubik's Cube, randomizing more than 100 physical parameters (including friction coefficients, object mass, and joint damping), and ultimately succeeded in transferring the policy to a physical robot.62% similarUnverifiedUnsloth can reduce GPU memory usage by approximately 70% compared to the standard Hugging Face training workflow.60% similarUnverifiedOpenAI Gym中绝大多数连续控制基准均基于MuJoCo构建58% similarUnverified在Hugging Face、OpenAI API、Google Vertex AI等平台成熟后，调用高性能预训练模型的门槛已降至只需几行Python代码58% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/539222API
curl https://kongchang.com/api/v1/knowledge/claims/539222MCP
get_claim(id=539222)