Show simple item record

dc.contributor.advisorAmes, Aaron
dc.contributor.advisorRogers, Jonathon
dc.contributor.advisorSawodny, Oliver
dc.contributor.advisorWohlhaupter, Uli
dc.contributor.authorWaters, Thomas Robert
dc.date.accessioned2018-01-22T21:14:32Z
dc.date.available2018-01-22T21:14:32Z
dc.date.created2017-12
dc.date.issued2018-01-09
dc.date.submittedDecember 2017
dc.identifier.urihttp://hdl.handle.net/1853/59310
dc.description.abstractPart 1 Driver assistance systems show the potential to increase the fuel economy and optimize the range of standard and electric vehicles. Eco-driving focused systems optimize velocity trajectories with respect to energy consumption and suggest these optimized speeds to drivers with the goal of reducing overall energy consumption. Because the systems have no direct control over vehicle behavior, the driver’s inclination to follow the commands is important to their effectiveness. This can be improved by personalizing the velocity commands to suit an individual’s driving behavior, requiring a model capable of accurately predicting styles of individual drivers. Two methods for identifying, modeling, and predicting driver behavior using driving data time-series are investigated. The first, pattern recognition-based approach breaks down the data into homogeneous segments using heuristic, dynamic programming, and bottom-up methods. Segments are grouped based on acceleration behavior and used, in conjunction with function-fit regression and system identification methods, to construct models describing driving behavior. Contrary to the first approach, the second, machine learning based method constructs a model using an entire time-series by analyzing relationships between multiple variables. Finally, each method is evaluated in it’s ability to accurately predict driver acceleration and velocity behavior. Part 2 Enforcing multiple, sometimes conflicting control objectives is a challenge present in modern advanced driver assistance systems. Drivers are capable of activating multiple modules simultaneously where safety must be guaranteed at all times. Examples includes adaptive speed regulation, where the vehicle must achieve a desired speed while maintaining a safe distance to any preceding vehicle, and lane keeping, where a vehicle is kept safely within the bounds of a lane. Provably safe algorithms for both adaptive speed regulation and lane keeping are introduced and used to run experiments on two robotic testbeds. The underlying algorithms are based on control Lyapunov functions for performance, a control barrier functions for safety, and a real-time quadratic program for mediating the conflicting demands between the two. The Robotarium, a robotic testbed that allows students, as well as researchers less experienced with hardware, to experiment with advanced control concepts in a safe and standardized environment, is compared with a more expensive OptiTrack based Khepera robot testbed.
dc.format.mimetypeapplication/pdf
dc.language.isoen_US
dc.publisherGeorgia Institute of Technology
dc.subjectDriver modeling
dc.subjectNonlinear control
dc.titleAdaptive driver modeling using machine learning algorithms for the energy optimal planning of velocity trajectories for electric vehicles and realizing simultaneous lane keeping and adaptive speed regulation on accessible mobile robot testbeds
dc.typeThesis
dc.description.degreeM.S.
dc.contributor.departmentMechanical Engineering
thesis.degree.levelMasters
dc.date.updated2018-01-22T21:14:32Z


Files in this item

Thumbnail

This item appears in the following Collection(s)

Show simple item record