PhD Internship - Remote

Who are we? To create an intelligent humanoid in the 3D world, the first step is to track real human motions in physical simulation. Although human motion capture and generation have been extensively studied in the past, it is still unclear how to let a humanoid track human motions precisely, while keeping the body balance at various terrains. Our goal is to explore this direction, and develop an effective cross-embodiment body and hand motion tracking algorithm in physical simulation. What you will be doing Creating a workflow to transfer motions from SMPL-X to humanoids (e.g. Unitree G1) in simulation. Developing an algorithm to perform cross-embodiment motion tracking. Evaluating the performance of the tracking algorithm. Combining generating motion models (e.g. diffusion models) to the tracking algorithm. Who you are Knowledge of PyTorch, deep learning, neural networks, C++. Knowledge of good software development practice, e.g. git, documentation. Knowledge of reinforcement learning and robot physical simulation. Knowledge of SMPL, SMPL-X, or character animation. At least one paper in CVPR/ICCV/ECCV/IROS/ICRA or equivalent. What we offer Research experiences both in academia and industry. Possibility of paper publication. 6-months working contract. Flexible remote work with global availability Our values Diversity isn’t just a statement at Meshcapade, it sits at the core of the company. We believe in the diversity of thought because we appreciate that this makes us stronger. Therefore, we encourage applications from everyone who can offer their unique experience to our collective achievements.