I'm Gbemi
Oloyede

Collaborated on building an autonomous robotic system capable of LiDAR-based navigation, shelf label recognition, and state-based decision-making. Developed components of the motor control and ROS 2 integration stack, along with a user interface for processing returns, monitoring stock, and issuing direct override commands when needed.

Developed a PPO-based reinforcement learning pipeline to train a drone racing agent in simulation and deployed the learned policy on a Crazyflie 2.0. Designed custom observation spaces, reward shaping, curriculum learning, and reset strategies to achieve fast, stable gate-to-gate flight.

Designed and implemented a complete VIO-based autonomy stack for a Crazyflie 2.0, enabling fully autonomous navigation without GPS. Integrated geometric feedback control, A*-based obstacle-aware path planning, smooth 3D trajectory generation, and visual-inertial odometry for real-time state estimation

This project focused on building an autonomous robot entirely from scratch as part of a collaborative team. From the beginning, we were responsible for every aspect of the system—mechanical design, electronics, and sensing. We designed custom parts in CAD and fabricated them using laser cutting, then assembled the robot and wired its electronics by hand. My work focused on integrating Time-of-Flight (ToF) sensors, which allowed the robot to detect obstacles and follow walls in its environment, building the web interface for monitoring sensor data and issuing direct override commands, developing I2C communication protocols, and a PID controller.