Connor Bossard

Projects

Develop a framework for planning under uncertainty against adversarial agents with unknown parameters (e.g., pursuit algorithms, maneuverability).
Integrate Expected Information Gain (EIG) into the cost function of trajectory optimization algorithms (MPPI, MPC, MBD) to promote active learning
Leads to control policies that "excite" the system to safely learn adversarial agent models while preserving mission objectives
Validate framework effectiveness in a high-fidelity, 6-DOF F-16 Dogfighting Environment

Developed a Transfer and Curriculum Learning Framework to train a robust navigation policy for adversarial environments
Pretrained an agent on expert demonstrations in a static maze using DAGGER and Behavior Cloning
Fine-tuned the policy for dynamic adversarial scenarios using Proximal Policy Optimization (PPO)
Achieved significant performance improvements over a baseline PPO policy in the adversarial environments
Link to full paper here

Implemented and compared non-linear adaptive control methods, including MRAC, Sliding Mode, and NN based approaches
Evaluated controller performance of each method on an F-16's ability to hold a trim condition with loss of elevator effectiveness
Link to full paper here

Performed Image Classification task on retinal images for automated disease detection
Engineered featured from pre-processed image data using Histogram of Oriented Gradients (HOG)
Implemented and benchmarked different ML architectures, including SVM, CNN's and YOLOv8 on classification accuracy
Link to full paper here

Fused TurtleBot Sensor Readings (LIDAR, Camera, Odometry, etc.) for state estimation in a maze environment
Implemented navigation and control algorithms for autonomous maze traversal
Fine-tuned a ResNet Image Classifier on few-shot learning on navigational signs to inform path decisions

Modeled and fabricated a robot consisting of 3D printed parts and heated inserts
Interfaced Arduino with three stepper motors and a resistive touch pad to determine ball’s location
Utilized inverse kinematics to determine how motor movement would impact platform orientation
Implemented a C++ based PID controller to balance a ball, using a resistive touch pad for orientation feedback and the motors for corrective action

Designed and modeled a compact electronics enclosure in CAD for integration with a lower-body exoskeleton
Fabricated the enclosure using 3D printers and heated inserts
Tested the wearable enclosure to ensure reliability against vibrations and environmental stresses