Kinematic Prior
Knowledge related to a robot's body structure, joint configuration and kinematic constraints that is independent of any specific task. Such priors are encoded as structural regularities in the weights of control networks, shaping how policies map observations to actions before task-specific training takes place.
Timeline (last 90 days)
The research team identified in the first-layer weights a structural regularity that stably reflects the robot's kinematic priors.
This kinematic prior appears under both reinforcement learning and imitation learning training paradigms, possibly reflecting common knowledge related to the humanoid embodiment rather than a product specific to a particular optimization process.
All Facts (2)
The research team identified in the first-layer weights a structural regularity that stably reflects the robot's kinematic priors.
50%UnverifiedThis kinematic prior appears under both reinforcement learning and imitation learning training paradigms, possibly reflecting common knowledge related to the humanoid embodiment rather than a product specific to a particular optimization process.
50%