Shake Bottle (Occupied)
Whether the policy can use self-generated rattling sound to distinguish occupied from empty bottles consistently across trials. The companion case showing the same active sensing behaviour under a different hidden physical state.
Why this task
The companion case showing the same active sensing behaviour under a different hidden physical state.
What it tests
Whether the policy can use self-generated rattling sound to distinguish occupied from empty bottles consistently across trials.
Context
Part of the real-robot evaluation of the HEAR framework, in which the primary metric is the sound-causal success rate: a trial counts as successful only if the robot completes the physical goal after the relevant acoustic condition has occurred. Reaching the goal early is recorded as a failure. All tasks were run on a Franka Panda with a single ambient workspace microphone and two RGB views.