Expired50% confidenceReleaseExact time
NVIDIA GEAR Lab, in collaboration with CMU and UC Berkeley, launched the Empire system, enabling AI coding agents to autonomously operate robots in real physical environments to complete experimental loops; the team officially claims the training strategy achieves a 99% success rate on multiple high-precision manipulation tasks, and the code will be open-sourced.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
7/8/2026
First Seen
This claim has expired. It is kept for the record and no longer describes the current state.
Valid until: 10/6/2026(expired)
Sources
Related Claims
UnverifiedSim-to-Real (simulation-to-reality transfer) has become the mainstream industry approach for humanoid robot motion control, with researchers using GPUs to run thousands of simulation instances in parallel in physics simulators to train reinforcement learning agents.73% similarUnverifiedA full-stack practical course simulating a Xiaomi Mall, using Cursor AI to automatically generate frontend and backend code combined with Figma AI for design, has been launched on Bilibili.72% similarUnverifiedThe Hongshuo AI R&D team has launched a practical AI programming in-house training program for enterprise R&D teams, with core tools focused on Cursor and Claude Code.71% similarUnverifiedCognition AI's Devin attempts to build a complete software engineering Agent environment, Adept AI's ACT-1 focuses on GUI operation automation, and OpenAI's Operator and Anthropic's Computer Use connect models to desktop control capabilities.70% similarUnverifiedAI研究者习惯使用Python和PyTorch/JAX等框架，而机器人工程师围绕C++和ROS节点构建实时控制系统69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/212923API
curl https://kongchang.com/api/v1/knowledge/claims/212923MCP
get_claim(id=212923)