Robot reinforcement learning

The operating task

Improve robot task policies against specified physical objectives and constraints. Observation, action and task completion are considered together within the selected embodiment and environment.

Softinator's contribution

Improve task behaviour against observable physical objectives and constraints in a controlled programme. The commissioned output includes policy candidate and task evaluation release.

Acceptance in use

Assess reward alignment and behaviour on representative conditions before drawing conclusions about transfer. The engagement defines the baseline, task scope and evidence used to assess the resulting contribution.