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. The company designs and operates global systems for data capture, data labeling, and hardware-enabled workflows used by leading AI labs and robotics companies to train and validate humanoid and embodied AI systems.
You will own the architecture for generating 3D objects from video, specifically reconstructing object geometry from egocentric footage of someone manipulating an object. Your work will ultimately produce simulatable assets with physical properties that can be dropped into physics simulators. This is a research role focused on an open, unsolved problem using real, large-scale data.
Key responsibilities include:
- Neural 3D reconstruction: developing models that recover object geometry from monocular/egocentric video using sequential and multi-view approaches robust to occlusion and low resolution
- Generative refinement: using generative 3D to complete and clean up partial reconstructions while rejecting hallucinated geometry
- Moving toward simulatable objects: progressing beyond geometry to assets with physical properties and rigging
- Contact-aware reconstruction: leveraging hand contact as a signal to constrain object shape and pose
- Deformables: tackling the frontier challenge of reconstructing objects that change shape during handling
- Prototyping and handoff: building research prototypes and transferring them to the integrations researcher for productization; publishing where appropriate
You will own the hardest, most open part of the objects effort—reconstructing objects that are small in frame, partly occluded by the hand, and sometimes deformable. You'll push the frontier, then hand results to the integrations researcher who takes them into the production pipeline.
Success looks like: manipulated objects reconstructing from egocentric video at usable quality (including small, occluded, and deformable cases); a credible path from interaction video to simulatable 3D object; research transferring into the pipeline through the integrations seat; and the objects pod building on the architecture and direction you set.
Tech stack: Python/PyTorch (primary) for model design and training; 3D deep learning with neural implicit/explicit representations and differentiable rendering; generative-3D toolkits and mesh processing (Open3D, trimesh); physics simulators for converting geometry to simulatable objects.
REQUIREMENTS:
- Strong 3D/neural-reconstruction research background: experience with neural implicit or explicit 3D, 4D reconstruction, or generative 3D at the frontier
- Deep learning depth: ability to design and train models, read and reproduce frontier papers
- 3D geometry foundation: knowledge of structure-from-motion, multi-view geometry, meshes, and differentiable rendering
- Research taste: ability to find tractable decomposition of hard, open problems and make steady progress
- Engineering capability: fast prototyping in clean Python/PyTorch; ability to get research ideas working on real data
STRONG PLUS:
- Generative 3D (diffusion/feed-forward 3D-generation families)
- Differentiable rendering, NeRF, or Gaussian splatting
- 4D/dynamic-scene reconstruction from video
- Physics simulation and rigging
- Hand-object interaction research; publications at top vision or graphics venues
