Issue glossary

Concepts & terms

The key ideas in this issue, explained briefly.

VLA: Vision-Language-Action model: maps visual observations and language instructions to robot actions.
Action chunk: A short sequence of future motor commands predicted in one policy invocation.
Flow matching: A generative learning method that transforms a source distribution into action trajectories.
UMI: Universal Manipulation Interface: a handheld interface for collecting demonstrations without a robot.
SLAM: Simultaneous Localization and Mapping: estimating pose while building an environment map.
Affordance: An object property describing which interactions are physically possible.
Proprioception: Internal robot state sensing, including joint positions and velocities.
Sim-to-real: Transferring a controller learned in simulation to physical hardware.
TCO: Total Cost of Ownership: acquisition, operations, maintenance and downtime costs.