The future of humanoid robots isn't JUST human: Haomiao Huang, Matter Venture Partners

In this talk

1. 1:10

   Why physical AI is gaining momentum

   Huang argues that AI’s impact depends on physical systems, from chips and data centers to robots that can help with health, daily tasks and building. He says progress in models, compute and demand has aligned: models help robots handle unfamiliar situations, edge compute supports real-time control, and needs such as manufacturing and data center construction create demand.

2. 9:01

   Humanoids as general-purpose machines

   Huang describes a humanoid as a general-purpose idea rather than one fixed body shape. A useful robot might need multiple manipulators, mobility, onboard power and compute, and the ability to work in human spaces; legs, wheels, size and limb count can vary by task. He compares this flexibility to a general-purpose CPU.

3. 12:59

   Reliability, data and components remain hurdles

   Huang says practical humanoid uses are still early, since 99% reliability at dishwashing could still mean a broken dish each week. He points to specialized work already underway, including Dexterity packing FedEx trucks and Path Robotics putting welding systems on mobile platforms. Scarce training data for demanding jobs such as structural welding, along with components like actuators, batteries, sensors and robot hands, are key bottlenecks.

4. 24:10

   Investment priorities and possible paths ahead

   Matter Venture Partners is looking across semiconductors, robotics, computing systems, manufacturing and logistics, energy, and AI for physical and life sciences. Huang expects more capable multipurpose robots in factories and warehouses, while home robots may first be useful when they handle around 80% of a task, such as putting away children’s toys. He says additional training data and improved hardware could move robots toward that level, while much more capable designs may take longer.