tech
Robot brain builders are pushing out of their GPT-2 era
Robot bodies are waiting for their AI brains to catch up.
TL;DR
- Physical AI is a hot investment sector, with companies raising billions for AI in robotics.
- Unitree, a leading robot maker, saw its valuation nearly halved after its IPO, highlighting the gap between physical capabilities and useful work.
- The Actuate conference revealed excitement and challenges in developing AI for robots, particularly the "robotics data crisis."
- Developing generalized robots remains difficult; current approaches focus on specific tasks or mimic advances in frontier AI labs.
- Autonomous vehicles are ahead due to data collection ease and avoiding physical manipulation, with their tooling influencing robotics development.
- Companies like Tesla, Wayve, and Uber are investing in robotics, leveraging their ML expertise.
- There's a debate on whether to co-design hardware and AI or focus on a general AI brain strategy.
- Task-specific robots are seeing field deployment (e.g., solar farms, industrial settings), while general-purpose humanoids remain in labs.
- Data management, especially visual and lidar data, is a significant challenge.
- The "ChatGPT moment" for physical AI is debated, with some envisioning consumer excitement and others, like Apple II or IBM PC moments.