Latty, A ROS based tricycle platform

Latty is a front-wheel-drive, non-holonomic tricycle robot platform built to explore, test, and benchmark path-planning and motion-control algorithms in simulation. The platform was designed as a modular research tool, enabling experimentation with localisation, mapping, and control strategies commonly used in autonomous mobile robotics.

To support simulation workflows, I created detailed URDF/Xacro model definitions compatible with Gazebo Classic, including wheel joints, steering mechanisms, collision elements, and sensor configurations. These models also integrate with ROS control plugins for realistic physics-based behaviour.

At the control layer, I developed lower-level interfaces for the platform using the ROS controller_manager, providing joint-state publishers, steering and wheel velocity controllers, and feedback pipelines for synchronizing simulation states with the ROS ecosystem.

On top of this, a dedicated ROS2 node was implemented to handle real-time joint control. The node publishes joint states, subscribes to command topics, and communicates with the robot model to update steering angles, wheel velocities, and platform kinematics.

To enable reliable motion tracking and navigation, Latty integrates a full suite of state estimation and localisation techniques, including:

• 1. Yaw estimation using Euler integration and exponential map formulations from IMU angular velocities.
• 2. Linear Kalman Filter (KF) and Extended Kalman Filter (EKF) fusion of IMU and wheel odometry for improved pose estimation.
• 3. LIDAR-based pointcloud processing for mapping and environmental understanding.

For visualisation and debugging, Latty uses ReRun, enabling intuitive playback and inspection of trajectories, sensor data, pointclouds, and state-estimation behaviour over time.

Overall, Latty serves as a versatile and extensible platform for mobile robotics research, combining robust simulation tools, sensor fusion, control theory, and ROS2-based software architecture.