This project will develop social behaviors for autonomous freeway driving. Human drivers signal their intentions to one another in various ways, including adjustments in speed and acceleration. We will refine previously developed intention prediction models and couple them with host-vehicle strategy generation methods to improve the performance and realism of autonomous vehicles interacting with other freeway traffic in a wide range of scenarios, with a primary emphasis on entrance/exit ramp handling and secondary emphasis on lane changing. The developed algorithms will be tested on CMU’s autonomous Cadillac SRX and, if possible, on relevant vehicles in GM’s continuing vehicle autonomy projects. Relevance to safety is twofold: use of such algorithms 1) in driver assistance systems to convey to drivers the intentions of surrounding cars; 2) in autonomous cars to provide less “robotic” and more natural, predictable, and socially comfortable driving.
Introduction This project will develop social behaviors for autonomous freeway driving. Human drivers signal their intentions to one another in various ways, including adjustments in speed and acceleration. We will refine our intention prediction models and couple them with host-vehicle strategy generation methods to improve the performance and realism of autonomous vehicles interacting with other freeway traffic in a wide range of scenarios, with a primary emphasis on entrance/exit ramp handling and secondary emphasis on lane changing. The developed algorithms will be tested on CMU’s autonomous Cadillac SRX and, if possible, on relevant vehicles in GM’s continuing vehicle autonomy projects. Problem statement Relatively little work has been done in creating autonomous driving behaviors that are socially acceptable. Drivers typically communicate or negotiate with one another through a variety of sometimes subtle clues. In earlier work, we have extended the rule-based behaviors approach used in the DARPA Urban Challenge race to a prediction-and-cost-function-based (PCB) approach that covers a wider range of scenarios. We have also performed initial work on prediction intention of other drivers (intention or iPCB), but this work is in its initial stages and has significant scope for improvement. Approach The proposed work will use Probabilistic Graphical Models (PGM) to probabilistically model and predict the intentions of other drivers. PGM is a powerful tool which allows coherent expression and straightforward simplification of the dependencies among random variables in complex systems. Initial results have enabled improvements over the iPCB method mentioned above. Further work will focus on extensions of the current PGM representation, followed by validation in simulation and on real vehicles. Tasks a. Refine current algorithms with better intention estimation using Probabilistic Graphical Models (PGM). b. Refine the PGM algorithm to handle multiple vehicles on an entrance ramp. c. Improve and analyze the PGM algorithm’s robustness to perception inputs and map errors. Define appropriate perception requirements. d. Collect human driving data in order to evaluate the realism of the PGM algorithm in relevant environments and tune its parameters. e. Test the PGM algorithm in the freeway driving domain with a focus on entrance ramp behaviors, but including also freeway distance keeping and lane changing. Improve and analyze the algorithm’s robustness to perception inputs.
Duration: 20 months (Jan. 1, 2017 – Aug. 31, 2018) High-level schedule: Aug. 31, 2017: Refined algorithm with better intention estimation via PGM Jan. 31, 2018: Human driving data collection July 31, 2018: Implementation and validation on CMU’s Cadillac SRX Aug. 31, 2018: Final report, documentation of hardware/software
1. Validate a basic capability to perform socially cooperative driving during the course of the proposed project. 2. Begin to transition this capability to GM R&D through the technology transfer mechanism already in place via the GM-CMU Autonomous Driving Collaborative Research Lab (AD-CRL). 3. The technology can then be matured through GM’s proven processes to the point that it may become suitable for deployment on passenger vehicles.
Accomplishments • Creation of a socially cooperative behaviors system with significant safety benefits via driver assistance and inclusion in autonomous driving algorithms • Validation of the system in simulation and on-road driving • Documentation of the results in a form appropriate for hand-off to a partner capable of full-scale deployment Metrics • Achieve 90% accurate prediction of neighboring drivers’ entrance ramp, distance keeping, and lane changing maneuvers • Achieve smooth and comfortable autonomous driving response to such maneuvers
Name | Affiliation | Role | Position | |
---|---|---|---|---|
jmd@cs.cmu.edu | Dolan, John | Robotics Institute | PI | Faculty - Research/Systems |
chiyud@andrew.cmu.edu | Dong, Chiyu | Robotics Institute | Other | Student - PhD |
No match sources!
No partners!