Searching and Tracking of Humans in Urban Environments by Humanoid Robots
Abstract
This thesis addresses the probabilistic challenge of searching for and tracking people in dynamic urban environments using mobile service robots. The problem is formulated as a "Hide-and-Seek" game, requiring the robot to plan actions under significant uncertainty caused by noisy sensors, limited field of view, and unpredictable human motion.
To solve this, the thesis proposes a comprehensive framework based on Partially Observable Markov Decision Processes (POMDPs). To handle the computational complexity of real-world state spaces, the approach utilizes Mixed Observability MDPs (MOMDPs) to separate fully observable variables (robot location, map constraints) from partially observable ones (human location).
Key contributions include:
- The application of Partially Observable Monte Carlo Planning (POMCP) to enable online, real-time decision-making in large environments.
- The integration of Particle Filters for robust state estimation and tracking.
- Extensions for cooperative multi-robot search, allowing teams to coordinate efficient coverage of urban areas.
The algorithms were validated through extensive simulations and over 3km of autonomous experiments with the Tibidabo humanoid robot in real-world urban settings in Barcelona, demonstrating robustness against dynamic obstacles and sensor limitations.
Related Materials
PhD Demonstrations & Experiments
- [Video] Hide and Seek with MOMDP (Section 4.10.1)
- [Video] Hide and Seek with POMCP (Section 5.9.1)
- [Video] Search and Track with POMCP (Section 5.9.2)
Master's Work (Pre-PhD)
- [Interactive] ALV Homing Simulation - Bio-inspired navigation method based on Average Landmark Vectors (Goldhoorn 2007)
Related Publications
Searching and Tracking People with Cooperative Mobile Robots
Autonomous Robots, vol. 42, pp. 1-25, 2017
A. Goldhoorn, A. Garrell, R. Alquézar, A. Sanfeliu
PDF | BibTeX | Project page
Searching and Tracking People in Urban Environments with Static and Dynamic Obstacles
Robotics and Autonomous Systems, 2017
A. Goldhoorn, A. Garrell, R. Alquézar, A. Sanfeliu
Continuous Real Time POMCP to Find-and-Follow People by a Humanoid Service Robot
IEEE-RAS International Conference on Humanoid Robots (Humanoids), Madrid, Spain, 2014
A. Goldhoorn, A. Garrell, R. Alquézar, A. Sanfeliu
PDF | BibTeX | Project page
Analysis of Methods for Playing Human Robot Hide-and-Seek in a Simple Real World Urban Environment
ROBOT 2013: First Iberian Robotics Conference, Madrid, Spain, 2013
A. Goldhoorn, A. Sanfeliu, R. Alquézar
Comparison of MOMDP and Heuristic Methods to Play Hide-and-Seek
International Conference of the Catalan Association for Artificial Intelligence (CCIA), 2013
A. Goldhoorn, A. Sanfeliu, R. Alquézar
Contact: alex (at) goldhoorn.net