Robotics Applied Scientist, Spatio-Temporal Scene Understanding & Memory

Investigate the appropriate representation for the robot's mission memory: a persistent spatio-temporal representation of what it saw, where, when, and how it changed, across missions and sites, so assets that move, disappear, or degrade are tracked rather than rediscovered. Build world models that let the robot reason and plan over that memory and recover physical parameters through interaction for physics-aware digital twins. Combine learned and physically grounded models, with geometry and dynamics as the ground truth. Bridge the perception-reasoning gap for long-term autonomy.

PhD in robotics, computer science, electrical engineering, aerospace engineering, mechanical engineering, or a related field, with publications at top venues such as RSS, CoRL, ICRA, IROS, T-FR, T-RO, CVPR, NeurIPS, or similar, on scene understanding, semantic SLAM, lifelong or spatio-temporal mapping, world models, or embodied reasoning. Hands-on experience with vision-language or foundation models applied to spatial reasoning, memory, or long-horizon embodied question answering. Depth in 3D representations: scene graphs, neural or Gaussian-splat maps, or object-centric maps, and the estimation underneath them. Strong Python; able to run your methods on real robots with C++ and ROS 2. Clear thinking about what a representation is for and how to evaluate it, and comfort with a fast-paced environment where research questions are set by what the next deployment needs.

Change detection or long-term mapping results from real deployments. Real2Sim, system identification through interaction, or digital-twin work. Experience with behavior trees or task planning driven by language models. Industrial or field robotics exposure.