Selected Projects

Research, coursework and personal projects.

Multi-Robot Coverage Planning on Fern Ridge Lake

Multi-Robot Coverage Planning

Colin Mitchell, Roger Wienaah, Julie Adams, Geoffrey Hollinger · OSU/SNC Wasp Project

Implemented a team-orienteering coverage planner with uncertain rewards in ROS 2 for coordinated depth search and contour tracking using sonar and acoustic communication. Built a high-fidelity Fern Ridge Lake environment in Unreal Engine for HoloOcean, simulated up to 50 uncrewed surface vessels, and ran field tests on the real lake with three BlueBoat robots.

ROS 2HoloOceanUnreal EngineBlueBoatMulti-robot
Multi-Robot Search Path Planning with ROS 2 Nav2

Multi-Robot Search Path Planning with ROS 2 Nav2

Manuel Agraz, Roger Wienaah, Michael Anku · Mobile Robotics (ROB 599) Project

Implemented a MILP-based multi-robot search planner to coordinate robots in capturing moving targets, and integrated it with Nav2 for waypoint navigation in Stage and RViz. Reduced average target capture time by 49% (140 s to 72 s) compared to exhaustive search across three environments.

ROS 2Nav2MILPPython
Auction-Based Multi-Robot Task Allocation and Path Planning

Auction-Based Multi-Robot Task Allocation and Path Planning

Clement Cantil, Roger Wienaah, Atharv Marathe · Sequential Decision Making in Robotics (ROB 534) Project

Implemented a dynamic auction-based task allocation algorithm for multi-robot coordination, using RRT, RRT*, RRT-Connect, and RRT*-Connect for collision-free planning and cost-based task assignment. Evaluated with up to 5 robots and 9 dynamic tasks, comparing task completion time and path costs.

Task allocationMotion planningPython
Friction-Aware Reinforcement Learning for Robotic Pick-and-Place

Friction-Aware Reinforcement Learning for Pick-and-Place Robots

Roger Wienaah, Jamie Santos and Dax Lynch · Learning Based Control (ROB 537) Project

Trained Truncated Quantile Critics (TQC) policies for pick-and-place with a simulated Franka Panda robot, comparing friction-aware state inputs, joint-level control, and curriculum learning. Curriculum learning improved low-friction grasping success by up to 20 percentage points with only 8% additional training.

Reinforcement learningPyTorchManipulation
Quadcopter Load Suspension

Quadcopter Load Suspension

Clement Cantil, Zane Meyer, Roger Wienaah · Linear Multivariable Control Systems I (ME 531) Project

Modeled a Crazyflie quadrotor with a cable-suspended payload in PyBullet, then developed and tuned a cascaded PID controller to reduce payload oscillations. Compared against an uncontrolled baseline, the controller reduced positional drift and improved flight stability.

ControlPIDPyBullet