Task 3 required that we navigate to a verbally-specified bottle using the same visual servoing approach as Tasks 1 and 2, then execute a bimanual manipulation sequence: 

• Step 1: the left arm grasps the bottle body from the side using a side-approach grasp pose computed from the bottle handle's bounding box and point cloud (red blocks)

• Step 2: the right arm grasps the bottle cap from the top using a separate grasp pose computed from the bottle cap handle's bounding box and point cloud (yellow block)

• Step 3: the right arm twists its wrist 180° (to unscrew the cap) and lifts to sleep pose (to remove the cap)

• Step 4: the left arm raises and rotates the bottle to pour out its contents (beads)

Task 2.99 successfully completed the first part of task 3. Using a speech request, TidyBot grasped the bottle using a side grasp pose. It also picked up the bottle, bringing it into stow pose.