14. RRT* LiDAR Local Planner

The JetRacer autonomous stack includes an RRT* (Rapidly-exploring Random Tree Star) based local planner. This module allows the vehicle to safely navigate around obstacles using LiDAR data, without requiring a pre-built global map or full SLAM integration.

Architecture

The rrt_star_planner.py node handles both the local occupancy grid generation and the path planning algorithm. It also includes a native Pure Pursuit controller to steer the JetRacer along the generated path.

graph TD
    subgraph Inputs
        SCAN[sensor_msgs/LaserScan]
        ODOM[nav_msgs/Odometry]
        TF[tf2_ros base_footprint]
    end

    subgraph rrt_star_planner Node
        GRID[Local Occupancy Grid Builder]
        RRT[RRT* Path Generation]
        PP[Pure Pursuit Controller]

        SCAN -->|Transform| GRID
        TF --> GRID
        GRID --> RRT
        RRT -->|Collision Free Path| PP
        ODOM --> PP
    end

    subgraph Outputs
        MUX[twist_mux Priority 6]
        RVIZ[RViz2 Markers & Grid]

        PP -->|cmd_vel_rrt| MUX
        GRID --> RVIZ
        RRT --> RVIZ
    end

How It Works

1. Local Grid: The planner maintains an egocentric (robot-centered) 2D occupancy grid, typically 6x6 meters with a 5cm resolution.
2. Obstacle Inflation: Laser scan points are plotted onto the grid and inflated by an inflation_radius (e.g., 0.35m) to ensure the vehicle body clears obstacles.
3. Goal Seeking: If no explicit goal is provided, the planner attempts to drive straight ahead (exploratory goal). If the straight path is blocked, it samples random reachable points in the forward hemisphere.
4. RRT* Execution: The planner grows a tree of reachable states, optimizing the path cost (distance) and respecting simple kinematic constraints (preventing turns sharper than 45 degrees).
5. Pure Pursuit Tracking: The integrated controller calculates an Ackermann steering angle to follow the generated path, scaling the throttle down when cornering sharply.

Usage

The RRT* planner is completely optional and disabled by default.

To start the planner on hardware:

ros2 launch jetracer_bringup autonomy.launch.py start_rrt:=true start_lane_following:=false

To run it in simulation:

# Terminal 1
ros2 launch jetracer_gazebo simulation.launch.py gui:=true use_rviz:=true

# Terminal 2
ros2 launch jetracer_bringup autonomy.launch.py start_camera:=false start_base:=false start_lidar:=false use_sim_time:=true start_rrt:=true start_lane_following:=false

[!IMPORTANT] The RRT planner operates at Priority 6 in the twist_mux. If both start_lane_following and start_rrt are true, the RRT planner will take precedence over the lane follower, allowing obstacle avoidance to override the centerline tracking.

Configuration Parameters

Parameters are centralized in src/jetracer_bringup/config/main_config.yaml.

Parameter	Default	Description
planning_frequency	10.0	Hz at which the planner generates a new path.
max_planning_time_ms	30	Maximum time allowed for RRT* execution to ensure real-time performance.
max_iterations	400	Maximum nodes added to the RRT* tree.
grid_resolution	0.05	Meters per cell in the local grid.
inflation_radius	0.35	Meters to inflate obstacles. Must be > half the robot width.
step_size	0.25	Distance between nodes in the RRT* tree.
pure_pursuit_lookahead	0.6	Lookahead distance for the tracking controller.
max_speed	0.3	Base forward velocity when path is straight.

Safety Mechanisms

The RRT module implements several safety fallbacks: 1. Stale LiDAR: If the /scan topic stops updating for > 1 second, the node publishes zero-velocity commands. 2. Blocked Path: If no path can be found after max_iterations or max_planning_time_ms, the vehicle halts. 3. Supervisor Override*: The safety_supervisor (Priority 255) and collision_assurance (Priority 8) remain active and can physically halt the vehicle if it gets too close to a wall, regardless of the planner's intent.