Human-like motion for robots that work with people.
Biomechanically accurate body models for seamless interaction.
Use Cases
Works in simulation, training pipelines, on-robot perception, teleop/IL, and safety/ergonomics tools.
Simulation & Training
Populate environments with realistic digital humans and motions, creating labeled datasets for ML pipelines to test perception, planning, and HRI.
Robot Perception & Teleoperation
Deploy synthetic data-trained models for human detection and pose estimation, retargeting human motion to robot kinematics for policy bootstrapping.
Safety & Ergonomics Analysis
Use accurate anthropometry and motion in workcell/digital-twin tools to check reach, clearances, and near-miss risk.
The Challenges & Our Solutions
Robots fail when they don't understand real human bodies. | Biomechanically accurate body models for seamless interaction.
The Challenges
-
Human perception & prediction is unreliable
Robots struggle to read human pose and forecast motion in real settings, leading to cautious or unsafe behavior.
-
Policy learning is slow and data-hungry
Collecting demonstrations and teleop sessions is costly, and transfer to the robot often breaks.
-
Real data misses long-tail edge cases
Rare behaviors and diverse body types are hard to capture at scale, limiting model robustness.
-
Contact and interaction are hard to validate
Evaluating grasps, balance, and human-robot contact requires realistic bodies and motions, not just stick figures.
Our Solutions
-
Human-aware perception & prediction
Train and evaluate models with anatomy-true digital humans (e.g., SMPL family) and realistic motion so systems detect, track, and anticipate people more reliably.
-
Imitation & teleop with motion retargeting
Retarget human motion to robot kinematics and generate consistent demo data to bootstrap IL/RL policies and improve teleop quality.
-
Synthetic training at scale
Create large, labeled datasets with diverse digital humans, clothing, and scenes (keypoints, masks, depth, contact) to cover edge cases safely and cheaply.
-
Contact-focused evaluation
Use accurate body/hand models and motion assets to test grasps, contact timing, clearance, and recovery behaviors before physical trials.