- Waterloo, Ontario, Canada
- Hybrid
- Full-time · Senior
- 5+ years
Advance the agentic drafting pipeline by designing, training, and evaluating vision language model agents to draft, check, and correct floor plans from panoramas and lidar, while leading research on scan alignment and contributing to 3D scene reconstruction.
Responsibilities
- Advance our agentic drafting pipeline: design, train and evaluate vision language model agents that draft, check and correct floor plans from panoramas and lidar.
- Build evaluation methods and datasets for agent accuracy, false alarms, cost and latency.
- Lead research on scan alignment as an initial project: recovering the relative positions of many 360° panoramas from 2D lidar and image data, robustly and at the scale of large buildings.
- Contribute to 3D scene reconstruction, depth estimation and camera-lidar fusion.
- Take your methods into production with our engineers, and publish or patent where appropriate.
Requirements
- PhD in computer vision, machine learning, robotics or a related field
- Equivalent research experience
- Nice to have: Spatial reasoning
- Nice to have: 3D understanding with multimodal models
- Nice to have: Reinforcement learning
