SlipstreamJobs tracks this role from the company's public career site. Apply directly on the employer's site.
Mecka AI is building the data infrastructure layer for robotics and embodied AI, designing and operating global systems for data capture, labeling, and hardware-enabled workflows used by leading AI labs and robotics companies to train and validate humanoid and embodied AI systems.
You will develop the full-body motion model that extends Mecka's current egocentric capture pipeline. Today, the system uses Meta's Momentum Human Rig (MHR) to fit hand and upper-body keypoints from head-mounted stereo/mono cameras, but the lower body is inferred and feet are approximated. You will build the learned model that replaces this approximation, conditioning on metric camera trajectory, hand and wrist poses, upper-body keypoints, 2D leg detections, and full-body ground truth from the ego-exo volume to produce accurate full-body pose in metric world coordinates.
Key responsibilities include: (1) Training generative motion models (diffusion, autoregressive, or masked) on mocap-scale and proprietary rig data, conditioned on partial egocentric evidence, to generate plausible lower body and feet when occluded; (2) Implementing physics-based tracking or contact-aware optimization to ensure physical consistency—ground contact, no penetration, no foot skate, plausible balance, including floor work (kneeling, sitting, lying, yoga); (3) Developing per-frame foot and body contact estimation with calibrated confidence; (4) Outputting poses native to MHR/SOMA-X with pose priors and constraints for production integration; (5) Collaborating with the audit team to define lower-body, feet, and contact metrics, commissioning motion captures to close named gaps; (6) Prototyping through to runnable production models and handing off to production owners; (7) Publishing research findings.
Success metrics include: lower-body and feet pass rate on the body cohort measured from Q1 and improving quarterly; physical-violation rate (penetration, foot skate, ground-plane errors) declining; floor work no longer breaking estimates; production solver integrating your prior and contact terms; public benchmark results and publication within the first year.
You will have access to hundreds of hours of ego-exo multi-view ground truth, thousands of hours of egocentric recordings, and mocap/IMU-suit sessions on request—data no public dataset has.
REQUIREMENTS:
- PhD or equivalent body of work in physics-based character animation, human motion generation, or 3D human pose and shape, with publications at SIGGRAPH/SIGGRAPH Asia, CVPR/ICCV/ECCV, NeurIPS/ICLR, or CoRL
- Demonstrated experience training diffusion, autoregressive, or masked motion models on mocap-scale data, conditioned on constraints (keyframes, end effectors, paths, text)
- Proficiency in motion imitation and tracking for simulated humanoids (DeepMimic/AMP/MaskedMimic lineage); comfort with Isaac Gym/Isaac Lab or MuJoCo and reinforcement learning at scale
- Knowledge of body models and kinematics: SMPL-X, MHR, or SOMA-X; retargeting; contact and collision modeling
- Clean PyTorch research code and ability to work with noisy, real, large-scale data rather than clean benchmarks
- Research taste: ability to decompose "full body from a head camera" into measurable steps and make steady progress
STRONG PLUS:
- Egocentric or partial-observation full-body estimation (head- and hand-tracking to body)
- Human–scene interaction and contact: scene-aware motion synthesis, volumetric body models
- Motion cleanup or denoising of estimated or corrupted motion
- Reusable motion priors for control, generative controllers, or large-scale motion tokenization
- Shipped research model into product or game engine; familiarity with Momentum/MHR tooling
TECH STACK: Python/PyTorch for model design and training; reinforcement learning at scale; Isaac Lab/MuJoCo; ProtoMotions- or MimicKit-class frameworks; SMPL-X/MHR/SOMA-X body models; AMASS-class motion capture.
