Portfolio
Autonomous Car Navigation with Mission-Level Planning
Overview
This project implements an autonomous navigation system for a car-like robot operating in a known indoor environment. The robot localizes itself, plans collision-free paths under non-holonomic constraints, and executes a multi-waypoint mission driven by time and resource constraints. Once initialized, the system runs autonomously without human intervention. The robot must visit selected locations that provide different benefits and return to a final endpoint before a deadline. Constraints on visit order and repetition require careful sequencing, highlighting the interaction between planning, localization, and control.
System Pipeline
The autonomy stack is organized as a modular pipeline:
- Localization: Particle-filter-based pose estimation using a known static map
- Motion Planning: Global planning in SE(2) using a Probabilistic Roadmap (PRM) with A* search
- Kinematic Constraints: Curvature-constrained planning for Ackermann steering (non-holonomic vehicle)
- Control: Velocity-based path-following controller for continuous trajectory execution
- Mission Logic: Task-level sequencing, constraint enforcement, and mission success evaluation
Planning is performed in a static environment with collision checking against known obstacles. The system does not use SLAM or dynamic obstacle avoidance.
Tools
- Python, ROS, MuSHR, Gazebo, RVIZ
Demo
