Sparse 3D LiDAR & Dynamic-Scene Planning
PhD research on programmable 3D sensing, dynamic object estimation, and adaptive motion planning.
Problem
Autonomous robots operating in dynamic indoor environments must perceive, track, and plan around moving obstacles using cost-effective sensors — often sparse 3D data from modified 2D LiDAR rather than expensive multi-beam units.
Approach
As part of my PhD at the IIT Bombay–Monash University Research Academy, I developed an end-to-end pipeline:
- Programmable 3D LiDAR — nodding 2D LiDAR with reconfigurable mirrors for dense 3D coverage and faster scan updates than conventional 2D LiDAR
- Two-stage dynamic object estimation — GMM-based real-time coarse tracking and shape estimation, followed by Kalman filtering for precise state estimation with arbitrary shape and motion primitives from sparse 3D data
- Adaptive BIT* — real-time motion planning among multiple dynamic obstacles in partially unknown environments
- Physics-based simulation — evaluation framework for planning under varying obstacle density
Results
- Publications in IEEE RA-L / IROS 2018 and Elsevier Mechatronics (2023)
- Presented at IROS 2018, Madrid (SDC travel grant)
- Related work on anytime motion planning (arXiv:1912.13461)
