TRACE: Efficient Transfer Learning for Neural Whole-Body Control
Overview
TRACE enables efficient cross-robot adaptation of neural controllers, achieving 96.08% success across diverse humanoid embodiments with minimal computational cost. It uses kinematic priors and reinforcement learning for motion transfer.
Generated from attributed reports · Updated 2 hours ago
Event evidence and corrections
0 attributed source owners. Ownership does not establish independent confirmation. Quantities are reported separately and are never added together.
Reported quantity · motion frames: 10000000 other · Basis not reported
“from Unitree G1 to Unitree H2 using 10 million motion frames and approximately eight hours of training on a single RTX PRO 6000 GPU, at less than 0.1% of the GPU-hour cost of training SONIC from scratch. The adapted controller achieves 96.08\% success on 67,953 evaluation motions under a height-scaled tracking criterion. Real-robot demonstrations span dynamic motion tracking and whole-body teleoperation, with collect”
Exact source · revision 1Source owner not reported
Reported quantity · training hours: 8 hours · Basis not reported
“from Unitree G1 to Unitree H2 using 10 million motion frames and approximately eight hours of training on a single RTX PRO 6000 GPU, at less than 0.1% of the GPU-hour cost of training SONIC from scratch. The adapted controller achieves 96.08\% success on 67,953 evaluation motions under a height-scaled tracking criterion. Real-robot demonstrations span dynamic motion tracking and whole-body teleoperation, with collect”
Exact source · revision 1Source owner not reported
Reported quantity · evaluation motions: 67953 other · Basis not reported
“from Unitree G1 to Unitree H2 using 10 million motion frames and approximately eight hours of training on a single RTX PRO 6000 GPU, at less than 0.1% of the GPU-hour cost of training SONIC from scratch. The adapted controller achieves 96.08\% success on 67,953 evaluation motions under a height-scaled tracking criterion. Real-robot demonstrations span dynamic motion tracking and whole-body teleoperation, with collect”
Exact source · revision 1Source owner not reported
Report timeline
Follow attributed reports and material updates.
- arXiv Robotics — research abstractsTRACE: Efficient Transfer Learning for Neural Whole-Body Control Across Humanoid Embodiments
TRACE is a framework for efficient cross-embodiment adaptation of neural whole-body controllers, preserving pretrained motion representations while specializing feedback control to the target robot. It reduces embodiment mismatch through kinematic priors and aligns joint dynamics, then combines encoder distillation with low-rank reinforcement learning on a compact motion set. The method achieves 96.08% success on 67,953 evaluation motions with minimal GPU-hour cost, demonstrating practical reuse of a pretrained humanoid controller across robots with different body scales and dynamics.
Timezone · UTC
Article dates follow your selected timezone. Briefing editions use Hong Kong time (UTC+8).