Scene Estimation

Team Cornell’s Local Map provides a list of obstacle states and measures of confidence to the path planner at 100 Hz, which, in the worst circumstances, can be used to avoid collisions safely. The Local Map intentionally does not take into account the constraints typical of normal driving, such as the road boundaries and structure, as other drivers on the road may not always follow these constraints. In normal driving scenarios, however, when all agents on the road obey the rules, the road constraints provide strong cues and additional structure to the environment. These cues can be used to improve the vehicle's estimate of its own location in the road network, as well as the location and threat level of obstacles in its nearby environment. These additional cues also allow the vehicle to plan in a simpler, more restricted world when all agents obey the rules of the road, ensuring more stable, robust response at the output of the path planner. It is the job of the scene estimator to identify and take advantage of these cues in order to enable the Tahoe to operate in this structured regime as often as possible.

The scene estimator acts as the interface between the sensors/Local Map and the intelligent planning hierarchy: it is the sensor fusion scheme that reduces the copious sensor data into the smallest pieces of information the path planner requires for normal driving. The scene estimator takes into account the constraints and assumptions of the Urban Challenge, whereas the Local Map and pose estimator make no simplifying assumptions.