Research Project

CoWBC: Learning Whole-Body Control for Unconstrained Cooperative Transport with Humanoid Robots

Submission for ICLR 2027 Submission Number: 15235

Rebuttal

Part 1 · Heterogeneous H2–G1 Collaboration

G1 can collaborate with a human partner or with H2, a humanoid with substantially different body dimensions.

Part 2 · Can TeamHOI Adapt to Cooperative Transport?

In our evaluation, TeamHOI could not be successfully adapted to our cooperative transport task.

Part 3 · Sim2Sim Contact Force Analysis

Contact forces on the object show how the robots support it during transport.

Part 4 · Why Use a Two-Agent Policy?

We test whether running two copies of a single-agent policy can produce coordinated transport.

Part 5 · New Real-World Demonstrations

Human–Robot Box Transport
Human–Robot Table Transport
Dual-Robot Table Transport
01 / 05

Human-to-robot transfer

Motion Retargeting

Human demonstrations are retargeted to two humanoid robots.

R—01

Carry the Box Sideways

00:16
R—02

Rotate the Box Together

00:10
R—03

Carry the Box Straight

00:10
R—04

Carry the Box Diagonally

00:22
R—05

Carry Straight, Then Rotate

00:14
R—06

Carry the Object Sideways

00:13
R—07

Carry the Box Sideways

00:17
R—08

Rotate the Box Together

00:12
R—09

Carry Straight, Then Sideways

00:16
R—10

Carry Sideways, Then Straight (L-Shape)

00:12
R—11

Carry Straight and Place the Object

00:11
02 / 05

Large-scale simulation

Policy Learning in Isaac Lab

The robots learn coordinated transport in parallel simulation.

T—02

Lift the Box Together

00:13
T—03

Rotate the Box Together

00:07
T—04

Balance and Carry the Box to the Target

00:08
T—05

Carry the Box Along a V-Shaped Path

00:14
03 / 05

Cross-embodiment coordination

Heterogeneous G1–X2 Results

Retargeted motions and learned policies for collaborative transport with a heterogeneous G1–X2 pair.

Motion Retargeting

Human demonstrations transferred to G1 and X2.

X—01

Cooperative Lift and Lower

00:10
X—02

Straight Carry and Turn

00:11
X—03

Sideways Carry, Lift, and Return

00:12
X—04

7-Shaped Transport

00:10

Policy Learning in Isaac Lab

Learned policies executed by the heterogeneous pair.

X—05

Straight Box Transport

00:10
X—06

Sideways Carry, Lift, and Lower

00:09
04 / 05

Pre-deployment validation

Sim2Sim Alignment

MuJoCo provides the final pre-deployment simulation for aligning the control stack with the real robots.

S—01

Lift and Lower the Box in Place

00:13
S—02

Carry the Storage Box Together

00:08
S—03

Rotate the Box Together

00:07
05 / 05

Physical experiments

Real-world Deployment

Hardware results for lifting, straight transport, rotation, and lateral motion.

H—01

Lift and Place the Box

00:08
H—02

Carry the Box Straight

00:12
H—03

Rotate the Box Together

00:13
H—04

Carry the Box Sideways

00:15