TurtleBot 2
Open source learning platform
The TurtleBot 2 is a differential-drive mobile robot built on a Kobuki base, designed as an open platform for learning and experimenting with ROS. I spent a lot of time with it over the years, first as a student, then later as a teaching assistant for the CS 7630 - Autonomous Robots class.
As a student
The class covered multiple layers of autonomy, from low-level control up to full mission-level behavior. The final project had the robot autonomously exploring the building floor while building a Wi-Fi signal strength map as it went, basically combining SLAM-based exploration with a secondary mapping task running on top.
I also used the TurtleBot 2 as a preliminary test platform for RL-based control, before moving on to outdoor tests with the Kingfisher. The setup used an AprilTag as a perception-based target for the policy to track. It’s a much friendlier place to debug a control policy: indoors, low stakes, and no risk of losing the robot in a lake.
As a fun side project, in collaboration with other students, we also got a small fleet of TurtleBots to follow each other in a daisy-chain, each one tracking the robot ahead of it using a unique tag attached to its back. Simple to set up, but it made for a fun demo.
As a teaching assistant
Later on, as TA for the same class (Spring 2024 and 2025), I helped keep the lab’s TurtleBot fleet running and supported students through their assignments. That meant a mix of hardware upkeep, software maintenance, and a fair amount of debugging help during office hours.
During that time, I worked on upgrading the sensor suite on the lab’s TurtleBots. In addition to the Kinect, we added an RPLidar for improved laser-based SLAM. That meant designing a 3D-printed mount for the new sensor, calibrating its extrinsics, and running integration tests to make sure everything played nicely together.
The improvement in map quality was remarkable. The lidar’s wider field of view and longer range compared to the Kinect’s depth sensor made a real difference, and it was enough to reliably run SLAM (via Nav2) across the entire second floor of the building — something that was a lot less consistent with the narrow field of view of the Kinect.
