5. Navigation and SLAM

The jetracer_navigation and jetracer_localization packages merge incoming geometric and dynamic signals to physically locate the robot within a room and chart the mathematically optimal track line over multiple rooms.

Sensor Fusion (The EKF)

A solitary sensor is untrustworthy. Encoders slip on dust (predicting infinite speed). IMUs succumb to electromagnetic drift when passing heavy wiring (distorting your sense of North).

Using the robot_localization ROS 2 package, we construct an Extended Kalman Filter (EKF) that mathematically averages the sensors simultaneously:

• Gyroscopic Yaw: Handled exclusively by the MPU9250 AHRS Sensor (fused onboard the MCU).
• Linear Drive: Handled exclusively by the Hall-Effect Wheel Encoders.
• Constraint Handling: We explicitly drop the IMU's raw Linear Acceleration vectors from the EKF array (ekf.yaml) because double-integrating them on 4GB compute introduces wild algorithmic runaway. Instead, we use raw angular velocity for rotation damping.

Autonomous 2D Pathfinding

graph TD
    subgraph Mapping Layer
      Lidar[RPLidar A1] -->|/scan| SLAM[slam_toolbox]
      SLAM -->|Creates| MAP[(2D Occupancy Grid)]
    end

    subgraph Nav2 Pathing
      MAP --> PLAN[SmacPlannerHybrid]
      EKF[EKF Localization] --> PLAN
      PLAN -->|Outputs Yaw Rate| NAVCmd[cmd_vel_nav]
    end

    subgraph Hardware Constraint
      NAVCmd --> INV[cmd_vel_to_steering.py]
      INV -->|Calculates Ackermann Angle| TWIST[twist_mux]
    end

Navigating Like a Car

Robots like Roombas use "Differential Drive"—meaning they mathematically output trajectories that demand turning in place. Your JetRacer uses a physical rack-and-pinion front axis. Therefore:

1. Nav2 relies on the SmacPlannerHybrid, configuring exclusively to an ACKERMANN motion model with a locked $0.40m$ turning radius.
2. Localization Stability: AMCL is configured with the nav2_amcl::DifferentialMotionModel, which provides the most stable approximation for car-like localization in Nav2 without demanding non-existent lateral motion sensors.
3. The cmd_vel_to_steering.py inverse kinematics script physically maps the requested Yaw rotation to your vehicle's physical $0.255m$ wheelbase, ensuring the tires actually pivot to the exact geometric angle requested by the AI.

Safety & Supervision

The navigation stack is supervised by the safety_supervisor node, which monitors LiDAR data for immediate obstacles and thresholds for wheel slip and battery health. If a critical fault occurs, the autonomous plan is immediately preempted by a hardware halt.

---

[!TIP] Next Step: Understand how we stop simultaneous sensors from overriding each other in 06. Behavior and Arbitration.