RoboBoom
Agility· September 15, 2026· 3:48

Agility Foundations: AI Stack

Jonathan Hurst, chief robot officer and co-founder at Agility Robotics, explains the software stack for humanoid robots operating in human environments.

In this talk

  1. 0:02

    Robots for human environments

    Hurst says human-centric robots are approaching a point where they can enter human spaces and do useful work. He frames enabling robots to perceive, reason, and act in unfamiliar environments as a central challenge for robotics.

  2. 0:51

    Build the stack from the hardware up

    Agility is building a software stack for embodied intelligence, which Hurst also calls physical AI. He describes its layers: capable hardware, coordination of 30 motors a thousand times per second for behaviors like walking and reaching, then skills such as picking up an object or opening a bin.

  3. 1:27

    Use examples to define skills

    Robot skills can start from demonstrations gathered through teleoperation, motion capture, video, or animations made in simulation. Hurst describes using these examples as a starting point for practice on the robot hardware.

  4. 2:03

    Combine skills and create robot training data

    A large language model can select and sequence available skills based on its interpretation of an environment. Unlike text or chess training, robot learning lacks a large existing dataset for choosing motor torques from sensor input in varied spaces, so that data must be created.

  5. 2:55

    Learn from deployments and prioritize blockers

    Hurst says perception has advanced to the point where robots can identify objects and understand their purpose. Agility's years of operating robots in real facilities have revealed practical challenges and changed priorities; he says the goal is machines capable of a wide range of manipulation and locomotion tasks.

Briefs summarise talks and interviews; timestamps open the video at that moment. Claims in a talk are the speaker's own.

More talks