ARTFEED — Contemporary Art Intelligence

DynaMAC: Multi-Agent Cooperation for Bimanual Robot Manipulation

ai-technology · 2026-07-27

Researchers propose DynaMAC, a lightweight framework for multi-agent robot cooperation that treats each arm as a dynamic task parameter, enabling sample-efficient bimanual manipulation without explicit leader-follower roles. The framework addresses causal limitations in dynamic environments where robot arms interact. DynaBench is introduced to evaluate these capabilities.

Key facts

  • DynaMAC is a policy-agnostic framework for multi-stream robot manipulation.
  • It treats the opposite arm as a dynamic task parameter.
  • It resolves causal assumptions that collapse in dynamic settings.
  • It preserves sample efficiency, computational speed, and flexibility.
  • DynaMAC does not require an explicit leader-follower relationship.
  • DynaBench is introduced to evaluate dynamic manipulation and bimanual coordination.
  • The approach is lightweight and policy-agnostic.
  • It provides a unified formulation for dynamic manipulation and bimanual coordination.

Entities

Institutions

  • arXiv

Sources