Senior Applied Scientist - Graph Optimization & Trace Alignment

The Road Features Group, a sub-organization within ADAS & ADS at TomTom, is the algorithmic engine that drives the creation of highly accurate HD maps to facilitate lane level navigation of Autonomous Vehicles. We create digital twins of road networks faster and more accurately than ever before. Our largest and freshest signal is crowd-sourced: millions of kilometers of vehicle traces and sensor observations collected from production fleets every day, alongside aerial and street-side imagery. Turning that noisy, massive stream into a precise lane graph is the core algorithmic challenge of the group. The Road Surface Graph (RSG) & Lanes team is pivotal in this effort, extracting drivable surfaces and lane centerlines at continental scale. Join us in setting new standards for mapping technology and making road feature extraction smarter and more efficient.

We are looking for a Software Engineer with deep experience in graph optimization, linear programming, and trace-based mapping (SLAM-style estimation) who can take large volumes of crowd-sourced vehicle trace data and turn it into a lane-level road graph. You will design the algorithms that align, filter, cluster, and fuse millions of noisy GPS and sensor traces, and that construct and optimize the resulting lane graph: combining traditional deterministic methods (linear and non-linear programming, factor graphs, combinatorial optimization on road networks) with modern AI/ML approaches where they beat the classical baseline. This is a role for someone who thinks in graphs and estimators first and treats both optimization solvers and learned models as tools in the same toolbox.

What You'll Do

Design and implement scalable algorithms that align, filter, and aggregate large-scale crowd-sourced vehicle traces (GPS, IMU, camera-derived observations) into consistent geometric evidence: map-matching, trace clustering, outlier rejection, and drift correction at fleet scale.
Build and optimize the lane graph itself: formulate lane-centerline extraction, connectivity, and topology inference as optimization problems (linear and non-linear programming, graph-based optimization, factor graphs / pose-graph techniques from SLAM) and solve them at production scale.
Combine deterministic and learned methods deliberately: use LP/ILP, dynamic programming, and probabilistic estimation where guarantees and explainability matter, and modern ML (learned detectors, GNNs, transformer-based sequence models over traces) where they demonstrably outperform: own the evaluation that decides which.
Architect high-performance implementations of these algorithms, making them robust, performant, and scalable for production-ready deployment across continental map scopes.
Collaborate with cross-functional teams to integrate trace-processing and lane-graph models into the production map-making pipelines, and feed measured failure cases back into algorithm improvement.
Stay current with the state of the art in trace-based mapping, crowd-sourced map inference, SLAM, and geospatial deep learning.
Deploy solutions using Docker containers on cloud platforms such as Azure.

What You'll Need

Master's degree in computer science, robotics, applied mathematics, Engineering, or a related field.
Software Engineer with at least 3-4 years of professional experience.
Strong software engineering skills, particularly Python; C++ experience for performance-critical solver and geometry code is a plus.
3-4 years of hands-on experience with optimization-based estimation: linear and non-linear programming, graph optimization (factor graphs, pose graphs, ILP on network structures), and probabilistic state estimation (Kalman/particle filters, MAP estimation).
Demonstrated experience processing large-scale vehicle trace or trajectory data: map-matching, trace alignment and registration, clustering, and sensor fusion from crowd-sourced or fleet sources (e.g., FCD/probe data, dashcam-derived observations).
SLAM and mapping knowledge: hands-on experience with SLAM or SLAM-adjacent pipelines (visual/lidar odometry, loop closure, bundle adjustment, HD map construction) applied to real sensor data.
Working knowledge of modern ML applied to geometric or graph problems (graph neural networks, learned lane/topology detection, sequence models over trajectories) and sound judgment about when a learned component belongs in a deterministic pipeline.
Experience with cloud computing platforms such as Azure and Databricks.
Proficient in deploying applications using Docker containers.
Ability to think end-to-end and deliver high-quality solutions.