This SBIR topic, Multi-sensor Detection and Tracking using Traversability Based Prediction, involves the sensing and prediction of moving entities while utilizing terrain features in order to reduce the false-alarm rate and to serve as a framework for accurate range and prediction horizon.
In order for fully autonomous robots to be integrated effectively into small combat teams, unmanned ground vehicles (UGV) must maneuver safely and intelligently among military ground forces and non-combatants. Detecting humans, both moving and stationary, is crucial for such safe operation and has been identified by the military as critical technology for deploying unmanned systems in theater. Detecting humans is a challenging problem due to the wide variety of positions and appearances which humans can assume. This problem is further complicated for small unmanned platforms because of weight, size, power, and cost constraints. Up to now, UGVs in theater have been used in a limited capacity:
- Teleoperation
- Small Size
- Slow Speed
These three characteristics strongly constrain the mission capabilities for ground robotic systems. The focus for this project can be broken into two main problems:
- Obstacle detection/tracking
- Prediction
To achieve robust obstacle detection capability, it is necessary to fuse information from various sensor modalities. Under various programs in the past, our team has developed obstacle detection/tracking capabilities using Ladar, LWIR, and visible light camera. We have started development on a multi-level approach to fusion that combines the information from the Ladar and vision sensors at both the feature and detection levels.
TACTIC, the prediction system, takes into consideration terrain and other traversability characteristics to generate a probability density function (pdf) using a planning structure. TACTIC drastically increases the efficacy of tracking behind occlusions and in highly cluttered terrain as opposed to more traditional approaches (e.g., Kalman filters).
This program is a DARPA Phase I SBIR started in 2009.
|
