Scalability of a Machine Learning Environment for Autonomous Driving Research

We study scalability of machine learning environments in the context of mixed collaborative driving. Mixed collaborative driving includes both human controlled vehicles and vehicles controlled by AI (Artificial Intelligence) that share the physical road resources (e.g., intersections and roundabouts...

Full description

Saved in:
Bibliographic Details
Main Authors: Debner, Anton, Hyyppa, Matias, Hanhirova, Jussi, Hirvisalo, Vesa
Format: Conference Proceeding
Language:English
Subjects:
Hybrid Systems
New Control Applications
Simulation
Online Access:Request full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:We study scalability of machine learning environments in the context of mixed collaborative driving. Mixed collaborative driving includes both human controlled vehicles and vehicles controlled by AI (Artificial Intelligence) that share the physical road resources (e.g., intersections and roundabouts). Many such driving situations cannot be easily created nor replicated in the real life. Therefore, development and testing of AI systems is often done with simulators.Machine learning environments must maintain a real-time understanding of their traffic situation. Scaling of the machine environment to multiple distributed nodes is required to support larger number of participating vehicles. Our experimental environment consists of the CARLA simulator, custom AI implemented with the TensorFlow framework, and a corner case search subsystem. With the corner case search subsystem we can automatically evaluate the AI in different driving scenarios. In this paper, we present how scaling of the environment to multiple distributed nodes affects its performance.
ISSN:2378-363X
DOI:10.1109/INDIN41052.2019.8972278