AI

Apple researchers use a Mac mini running Qwen locally to control a Unitree G1 humanoid robot. Trained on 1,000 simulated homes, it autonomously sits, crawls under tables and navigates unfamiliar rooms, achieving 91% success across 360 real-world tria

Read full article at source ↗

Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 08/11/2026, 20:47:26 UTC.

Briefings

Apple researchers successfully deployed a Mac mini running the Qwen large language model to control a Unitree G1 humanoid robot. The system autonomously navigated 360 real-world tria, achieving 91% success rate. The robot was trained on 1,000 simulated homes and demonstrated abilities such as sitting, crawling under tables, and navigating unfamiliar rooms.

Key points

• Apple researchers used a Mac mini running the Qwen large language model to control a Unitree G1 humanoid robot.
• The system achieved 91% success across 360 real-world tria.

Context

This demonstration shows a practical application of large language models in robotics, where a consumer-grade Mac mini can operate a humanoid robot in complex, unstructured environments.

Limitations

Evidence is from a Reddit post; specific training data details are unverified. The 91% success rate applies only to the 360 tria tested. No causal claims about the model's generalizability or future capabilities are supported by this evidence.

Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 08/11/2026, 20:51:55 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.

View archived briefing on IPFS
bafybeibkof3nqmibtziuemgramg7ty7htnngew6var5daejtan7zxsixyi