Cognitive roboticsDynamic manipulationOperational intelligence

DyNAMO

DyNAMO advances cognitive robotics by bridging perception, reasoning, and action — an integrated framework for dynamic navigation and manipulation, carried from simulation through to real hardware.

A cognitive robotic system built to adapt to unstructured environments and learn through interaction. It pairs foundation-model reasoning with classical navigation and control to reach versatile, robust behavior in real-world settings.

Objective

Integrate navigation, manipulation, and reasoning

Dynamo demonstrates an integrated system where a humanoid robot identifies, grasps, and places objects while working with an autonomously navigating mobile base. The core challenge is Sim2Real transfer: behavior developed in simulation must survive the noise and constraints of physical operation.

Research question

Can navigation and manipulation developed in simulation transfer to real hardware as one integrated system — and how reliably does each subsystem hold up under physical conditions?

RidgebackUnitree G1Isaac LabNav2

Experiment log

Milestone 12Zero-shot hand transfer

Towel pick and place on three-finger hands

The same pipeline runs on a soft, deformable object, a towel, with the G1 fitted with three-finger Dex3 hands. The simulation only ever modelled two-finger hands, so the policy has never seen this hand. The clip is shortened and sped up 1.5 to 2 times.

Outcome·The G1 picks the towel up and drops it into a box with no fine-tuning for the new hand: a zero-shot transfer to three fingers. It does not hold for every task. The cube task does not transfer zero-shot to the three-finger hand.

12/12

Milestones

12/12

Zero-shot hand transfer