Ground Robotic Research, LLC

Recent testing at Fort Hood, TX showed state-of-the-art autonomy for critically needed applications for the US Army.

Unmanned operations in snow and rain.

Owners Karl Murphy (L) and Alberto Lacaze (R) in front of an unmanned Stryker vehicle.

Three Dimensional Dynamic Environments Path Planner (3DDEPP)

This SBIR topic, Three Dimensional Dynamic Environments Path Planner (3DDEPP), will allow for an enhanced use of three dimensional search methods for use in obstacle avoidance in crowed environments.

Planning with dynamic entities is at the core of the "safe ops" risk as identified by FCS' Autonomous Navigation System (ANS) and other robotic programs. Autonomous robotic systems will not be extensively deployed unless they can safely operate in environments sharing spaces with other dynamic entities.

One of the biggest challenges is that most planning systems do not understand the idea of a dynamic entity. They represent static entities in a grid map based representation. Maps are an important tool to reduce the false alarms that current sensing systems provide. By accumulating the detections or lack there of in a particular cell it is possible to improve the classification of static obstacles. This technique is not well suited for environments with movers. Much of the research in detection of movers ignores the fact that this simple technique is no longer straight forward in dynamic environments. A system that represents movers will also have to cope with the statistical disadvantages that may be caused by not being able to simply collect detections at one cell as the content of the cells can change with time.

Planning in dynamic environments requires utilization of a 3D planning space (x,y,t) for ground vehicles and a 4D space for aerial vehicles (x,y,z,t). These search spaces are larger than the typical 2D (x,y) spaces used both for ground and in general for aerial vehicles where the altitude changes are mostly used as constraints rather then full dimensions of search.

In this project, we present an innovative mechanism for predicting the motion of other entities that intertwines the planning and the prediction tasks. In previous systems, the prediction was decoupled from planning. However, in dynamic environments our motion affects the motion of other entities. Even more importantly, our future motion affects the future motion of other entities. Therefore, there are multiple possible futures depending on our actions. The space of search is greatly increased by considering this interaction. Our system minimizes the computational complexity of prediction by merging the process of prediction and planning.