Invited Talk: Motion control bits for homemade robots by Prof. Stéphane Caron

In this talk

1. 1:32

   Small robots for focused research

   Caron draws on work with humanoids, quadrupeds, and Appy, a roughly five-kilogram wheeled biped. He describes pairing quasi-direct-drive actuators with 3D-printed parts to build and repair robots quickly. Smaller, task-specific platforms can make experiments easier to repeat and reduce the reluctance to test hardware that might fall or break.

2. 14:28

   Sharing robotics software

   Caron separates motion control into making a robot move and defining the task it should perform, then describes specialized tools alongside recurring engineering needs such as logging, packaging, and build systems. He advocates sharing research software through GitHub and demonstrates a Python package for loading robot descriptions, including models that can be visualized with Pinocchio.

3. 21:08

   Python and C++ in a control pipeline

   For performance-critical work, Caron uses C++ alongside Python, but says bindings can make prototypes cumbersome to change. His Voolp approach exchanges observation and action dictionaries between a faster C++ process and a slower Python controller through inter-process communication. In a Raspberry Pi experiment, Python ran a 200 Hz control loop with comparatively low timing variation while handling inverse kinematics and balance computations.

4. 31:56

   Balance frequency and hardware feedback

   Caron describes divergent component of motion feedback for balance: a biped shifts its zero moment point, while a wheeled robot moves its wheels to adjust the centroidal moment pivot. He cites research indicating that balance performance can tolerate slower sampling below a system-specific threshold; for HRP-4, he gives a minimum of 16 Hz. In discussion, he says the right frequency depends on the disturbances the robot must handle, and favors real hardware time for controller development while using simulation to explore unfamiliar morphologies.