General robotic foundation models will unlock Cambrian explosion of robot applications
Just as PCs and LLMs enabled anyone to build software applications, a general-purpose robotic foundation model will let anyone build robot applications without solving the intelligence stac…
Multimodal LLMs provide common sense for robotics long-tail scenarios
Multimodal LLMs contain world knowledge that can be grounded in physical situations via chain-of-thought reasoning, solving the 'common sense' bottleneck for handling novel scenarios. This…
Robotics data flywheel starts at 'useful enough' not 'perfect'
The key to solving robotics data scarcity is not pre-collecting massive datasets but deploying systems that are useful enough to operate in the real world, where they autonomously gather di…
One intelligence problem across all robot form factors
Physical intelligence is embodiment-agnostic: the same foundation model can control humanoids, arms, bulldozers, or drone swarms. The fundamentals of physical interaction, causality, and ob…
Combining generative AI knowledge with RL superhuman performance is the grand challenge
Generative AI (LLMs) captures human knowledge but mimics human performance; deep RL (AlphaGo) discovers superhuman strategies but lacks world knowledge. Robotics needs both: web-scale knowl…
ML flips Moravec's paradox: data-rich physical tasks become easy
Tasks easy for humans but hard for robots (dexterity, manipulation) become tractable with ML when data collection is straightforward. The remaining hard problems are where data is scarce an…
Robot improvement bottleneck shifted to semantic reasoning — solvable by language coaching
Robots now fail at task interpretation, not low-level control. Adding semantic labels (coaching) to autonomous experience improves generalization without new teleoperation data. This means…
Humanoid locomotion uses simulation; manipulation uses real data — winner unclear
Two divergent paradigms exist: humanoid acrobatics rely on heavy simulation with near-zero real data, while manipulation relies on massive real-world datasets with little simulation. The fi…
Robot hardware costs plummeted 100x enabling general-purpose research
Robot arm costs dropped from $400k (PR2) to ~$3k per arm due to learning-based control that works with low-precision hardware. This hardware affordability constellation makes general-purpos…