Carnegie Mellon University

Off-Road Autonomy

NREC developed an autonomous navigation system for large scale, unmanned ground vehicles (UGV) operating autonomously in a wide range of complex, off-road terrains. Even when navigating the most complex terrains, the navigation system running on the Crusher vehicles can drive for 12 kilometers, on average, before needing help from a human operator. The program ended a few years ago, but Crusher still remains the most capable autonomous ground vehicles for complex, off-road terrain.

Navigating complex terrain at speed and with minimal human supervision is a major challenge for UGVs. The need to recognize obstacles requires a dramatic improvement in perception capability. Also, the continued likelihood of running into obstacles requires a vehicle that is rugged enough to continue operating after sustaining tolerable damage in collisions.

Mission Planning
Our team developed an advanced mission planning system that can take advantage of overhead data such as imagery and elevation maps. With such information, the mission planner can estimate the best path for the vehicle given the mission parameters. The mission planner can find the best route to get to destination more rapidly or to stay hidden from specific observers, for example. Machine learning also helped optimize the performance of the mission planning system. With human examples of the best route to take to remain concealed, the mission planner can leverage that learning and use it to derive the optimize route when given a new map or terrain.

Perception
A combination of ladar and camera systems allow the vehicles to dynamically react to obstacles and travel through mission waypoints spaced over a kilometer apart. The use of overhead data via terrain data analysis can be utilized for global planning. These vehicles analyze, plan, and execute mobility missions over extreme terrains without any human interaction at all.

UPI’s advanced perception and autonomy systems provides unmatched automation capabilities at higher speeds. They feature machine learning technologies that enable Crusher to learn from sensor inputs and autonomously navigate complex, unfamiliar terrain.

Autonomous Navigation
With the information about its surroundings provided by the perception system coupled with the pre-planned route from the mission planner, the navigation system finds the best path to get to destination given the obstacles and the details of the terrain that were not part the overhead data. Several times per second, the navigation system makes new plans and evaluates the best path given the environment, mission parameters and capability of the vehicles. Even on the most difficult and complex terrain, Crusher can find its way to the destination and can navigate for several kilometers without any human help.

Field Experiments
UPI experiments encompass vehicle safety, autonomous navigation, the effects of limited communications bandwidth and GPS infrastructure on vehicle and autonomy performance, how effective is autonomous navigation when driving over complex off-road terrains, and how vehicles and their payloads can be effectively operated and supervised. By showcasing Crusher's unique terrain capabilities and advanced sensing and autonomy systems, these field tests influenced the development of autonomous vehicles.

Off-Road Autonomy Field Tests

Photos

NREC's off-road autonomy
NREC's off-road autonomy
NREC's off-road autonomy