Eric Berndt Email and Phone Number

• Built deep learning perception/decision making models for 170mph L5 autonomous IndyCar, 4-car race in Indy Autonomous Challenge• Designed instance segmentation model in PyTorch & CUDA, then quantized model in TensorRT & C++, achieving 250Hz on board• Building reinforcement learning adversarial multi-agent path planner and opponent vehicle motion prediction models for decision making• Built semi-supervised auto-labelling pipeline using SAM & temporal tracking model to label 500,000 images from our 100TB database• Developed sensor fusion, multi-car tracking system with LiDAR, radar & GPS to track rival cars using extended Kalman filtering & ROS2

• Researching generative path planning using imitation learning with Diffusion models to fly adversarial autonomous drones• Utilized visual-inertial odometry & depth estimation pipeline with stereo cameras & IMUs for state estimation onboard with Nvidia Jetson• Designed & built hyper maneuverable tilt rotor UAV platform and trained policies in AirSim simulator for DARPA AI program

• TA for EECS c106A (Intro to Robotics), taught sections, lectured, authored robotics guide & course material

• Developing soft-body physics race car model in simulator to for sim-to-real transfer learning• Creating machine learning pipeline to train car from simulator

• Lead 7 person team developing product recommendation program and mentored Comp Sci students

• Created SQL database for inventory & shipping/receiving for multiple clients in a storage facility• Designed UI to meet company specs & pitched software to CEO, who chose to adopt software