Probabilistic Tracking and Trajectory Planning for Autonomous Ground Vehicles in Urban Environments
Abstract
The aim of this research is to develop a unified theory for perception and planning in autonomous ground vehicles, with a specific focus on obstacle tracking/identification and trajectory planning, so as to enable reasoned, intelligent planning rather than simple reactive planning. This final report details our contributions. First, we developed a new method for anticipating obstacle motion using Gaussian Processes in order to enable contingency planning. Second, we developed three new mapping strategies including a unified terrain model based on a Markov Random Fields; soft relative maps; and fusion of stochastic maps. Third, we developed a methodology for capturing negative information (e.g. reasoning about areas where there is no sensor data), which is then used to improve tracking of obstacles. Fourth, we developed a new smoothing method which enables real time update of complex density functions (e.g. complex maps, tracking dynamic obstacles) in the presence of sparse data. We have published /accepted 23 journal articles and conference papers.
Document Details
- Document Type
- Technical Report
- Publication Date
- Mar 05, 2016
- Accession Number
- ADA635945
Entities
People
- Mark E. Campbell
Organizations
- Cornell University