Position, Velocity, and Attitude Determination

In designing the Tahoe’s position, velocity, and attitude estimation system (the pose estimator), Team Cornell considers three main objectives. First, the pose estimator must have inplane accuracy at the meter level if the vehicle is to cross checkpoints, find parking spots, stop at stop lines, and navigate in environments where visual or other sensory cues might be weak or unavailable. Second, the pose estimator must be able to provide accurate estimates of differential vehicle motion, such as filtered angular rates and velocities, to allow external sensing systems to estimate various quantities in the dynamic environment from a vehicle-fixed coordinate frame. Finally, and perhaps most importantly, the pose estimator must be robust against the challenges of the urban environment. In particular, it must be able to withstand reasonable levels of satellite visibility and occlusion, as well as short blackouts and signal corruption in the urban canyon, without providing brittle or biased estimates of vehicle position. These three design objectives have driven the development of Team Cornell’s pose estimator. Most important in the early development process was Team Cornell’s decision to design and build a pose estimator in house rather than buy an off-the-shelf alternative.

Obstacle and Environment Sensing

The Team Cornell sensing system is designed around the challenging requirement of being prepared to see everything in an unknown urban environment. This broad requirement is further subdivided into the ability to detect three types of environmental features: static obstacles, moving obstacles, and various types of roads. Most importantly, Team Cornell has evaluated and selected its sensors based on their capability of detecting these three aspects of the environment, not on their ability to track these objects independently. In this way Team Cornell adopts a purely Bayesian approach to sensor fusion, preferring to fuse raw sensor output in a centralized estimation scheme rather than depend on individual sensors’ proprietary tracking algorithms. The sensor suite includes three IBEO 1.5D LIDARs, five 1D SICK LIDARs, five millimeter-wave radars, and four cameras.