Robots Among Us: Advances in AI for Everyday Androids
summary
Claims a working pipeline: train agents in Isaac SIM (they run ~4,000 simulated robots, show progress at ~100 and ~300 epochs), fine‑tune a robotics foundation model (examples: Groot 3B with a 2.1B VLM or Pi 0.6) on cheap hardware, deploy trajectories to a Unitree G1 and small SO101 arms, and perform tasks like sorting dice and folding laundry. Tools: reinforcement learning, SLAM, motion planners, Nvidia GPUs/Jetson; demo mostly succeeds (one cable failure), market estimate cited as multi‑trillion.
AI-written summary · about
chapters (18)
- 0:00Welcome
- 1:48Talk agenda
- 2:36Robotics history
- 10:24Why robotics advances now
- 16:55Robot components
- 18:58Actuators explained
- 20:04Sensors and perception
- 21:08Control systems
- 21:51SLAM algorithm
- 25:23Motion planning
- 28:38Reinforcement learning
- 33:24Simulation to real-world
- 33:44Foundational models
- 36:56Nvidia Groot training
- 38:46Fine-tuning models
- 40:54Live demo
- 44:02Applications and future
- 46:13Q&A
related talks
32mThe Road to General AI
16mReal-World Robotics as the Next Frontier for AI?
19mThe Physical Turing Test: Solving General Purpose Robotics
21mDemocratizing AI with Open Source Multi-Agent Systems
42mComputer-Use Agents: From SFT to Classic RL
35mThe Fight for the Future of Artificial Intelligence
47mBeyond Agents: What AI Strategy Really Needs in 2025
14mReliable Agentic AI, the Open Way