Autonomy |

Rocker Robotics — Intelligent Ground Vehicle Competition

Wed, 01 May 2024 00:00:00 +0000

Autonomy and Controls Technical Lead (2022–23), then team member (2023–24) for South Dakota Mines’ entry in the .

Results: Won Rookie of the Year award and placed in the design competition.

Technical contributions:

ROS-based sensor integration (LIDAR, GPS, camera)
Lane detection via computer vision
Path planning with A*
Trajectory tracking with Model Predictive Path Integral (MPPI) Control
Frontier-based autonomous exploration for course navigation

Autonomous Spacecraft and Sensing

Mon, 01 Apr 2024 00:00:00 +0000

A systems-level demonstration of autonomous spacecraft operations, presented to a general audience and built entirely in the astrodynamics simulator.

Capabilities demonstrated:

Attitude correction and momentum management
Onboard debris detection via simulated optical sensor using an optical flow background removal scheme
Autonomous debris avoidance via thruster actuation
Cyber-attack simulation targeting autonomous operations with minimal detectability
Anomaly detection using LSTMs and isolation forests

Autonomous Vehicles with Safety Guarantees

Thu, 01 Feb 2024 00:00:00 +0000

A demonstration suite for safe autonomous vehicle trajectory optimization.

Lane, vehicle, pedestrian, traffic light, and sign detection pipelines
Autonomous lane change via nonlinear model predictive control (nMPC)
Safety-guaranteed lane change using Control Barrier Functions integrated into nMPC (CBF-nMPC), with formal guarantees of non-collision with simulated traffic

Detection of Debris from a Dynamic Satellite Platform

Thu, 01 Jun 2023 00:00:00 +0000

B. Outland, R. Loveland

Developed a novel debris detection algorithm for space-based optical sensors. Classical background-subtraction methods systematically remove faint, slow-moving objects — exactly the objects of interest in space domain awareness.

This approach uses an optical flow background removal scheme specifically designed to retain faint objects, followed by object detection classification. The method was validated on simulated imagery from a dynamic (tumbling/maneuvering) satellite observer.