Research Scientist, SLAM & VIO Location: New York (in office) ·

Type: Full-time ·

Openings: 1 ·

Salary: US$200K – $250K equity The role

Build the next generation of spatial and temporal intelligence. Our client's perception team already handles classical state estimation and VIO; this role architects and trains proprietary foundation models from scratch. You'll build in-house equivalents to leading 3D reconstruction architectures and robust optical flow models for egocentric vision. You'll train on a large, continuous stream of proprietary ground-truth data and lead on new perception problems as they emerge.

What you'll do

- Design, implement and train feed-forward and per-scene differentiable optimization architectures for 3D geometry extraction
- Scale multi-view architectures across multi-GPU clusters on massive multi-modal spatial datasets
- Develop novel loss functions and attention mechanisms tailored to proprietary data distributions
- Build optical flow models that handle extreme motion blur, rapid rotation and sudden occlusion from head- and robot-mounted cameras
- Use flow to segment dynamic actors, track objects through manipulation and regularize downstream world models




- Rapidly prototype models for tracking, segmentation and multi-modal sensor fusion
- Connect model outputs to large-scale dense reconstruction pipelines (3D Gaussian Splatting, NeRFs)

What you bring

- Deep expertise in deep learning, 3D computer vision and temporal/video modeling
- Proven experience training large-scale vision models from scratch, beyond inference or fine-tuning
- Strong grasp of modern feed-forward 3D networks and dense motion estimation
- Mastery of PyTorch and deep learning scaling frameworks
- Experience curating multi-terabyte image and video training datasets
- Comfort in a fast-paced environment where priorities shift with new research or hardware

Strong signals

- First-author publications at CVPR, ICCV, ECCV or NeurIPS on 3D deep learning, optical flow, video generation or spatial transformers
- Work with egocentric video datasets (e.g., Ego4D, Ego-Exo4D)
- Custom CUDA kernels for 3D operations, ray marching or correlation volumes

Please note: Research Scientist roles require highly specific expertise. Apply to only one research role; you may also apply to non-research roles.

📌 Research Scientist, SLAM & VIO - ON-SITE (Canada)
🏢 Targeted Talent
📍 Canada