Coordi-pi: Coordination-Aware Dual-Hand Action Representation for Bimanual Dexterous Manipulation

Key points

• Existing bimanual VLA policies commonly concatenate right- and left-side arm--hand actions at each control timestep, leaving cross-hand interactions implicit within a single action vector.1
• We introduce Coordi-$π$, a structured action representation for bimanual dexterous manipulation.1
• Coordi-$π$ represents the right- and left-side arm--hand actions as individual action tokens and organizes them into an interleaved sequence, explicitly exposing the two sides within the action sequence.1
• To explicitly encode hand identity and temporal synchronization in this interleaved representation, a learnable hand-identity embedding distinguishes the two sides, while sequence--control-timestep factorized rotary positional encoding (RoPE) separates sequence position from control timestep, allowing actions at the same control timestep to share a temporal coordinate while retaining distinct sequence positions.1

Sources (1)

• [1]Coordi-pi: Coordination-Aware Dual-Hand Action Representation for Bimanual Dexterous Manipulation
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 14, 08:08 AM

  Vision-language-action (VLA) models have shown strong potential for general-purpose robotic manipulation, but extending them to bimanual dexterous manipulation introduces a high-dimensional and structured action space.

  Existing bimanual VLA policies commonly concatenate right- and left-side arm--hand actions at each control timestep, leaving cross-hand interactions implicit within a single action vector.

Extractive summary: sentences quoted from the sources.