SVTRIP (Stochastic Vehicle TRip Prediction) generates naturalistic vehicle speed profiles following training on a large datasets of recorded driving data.

From a route to a speed profile: the end user defines a trip on a digital map (e.g. HERE maps) or a travel simulator (e.g. POLARIS), by providing origin, destination, and waypoints. SVTRIP then extracts a macroscopic definition of the trip with attributes such as speed limit, travel time, road class, and intersection type for each segment of the trip. In the last stage, SVTRIP sequentially generates a time-indexed speed signal that fits the attributes of each segment.

SVTRIP Route To Cycle
Fig.1 - Route To Cycle.

Data-driven. SVTRIP models how people drive by training on very large databases of recorded driving data. Every transition in speed that occurs in the output has been observed in the real-world dataset. We use a combination of Markov chains and machine learning to train on thousands of hours or real-world driving.

Naturalistic and Stochastic. The algorithms behind SVTRIP include the randomness inherently associated with human driving. As a result, the output of multiple runs/generations of SVTRIP with the same route target as an input results in a different result each time. The result is naturalistic, because it is made of speed transitions that actually occurred in the training dataset.

Route-specific. Each speed profile SVTRIP generates is particular to the route provided as input. For each segment of the route, the algorithm identifies the most relevant subset of data (e.g., high-speed highway driving) and generates a speed trace whose attributes match the target attributes of the segment (e.g., segment distance, initial speed, travel time, speed limit, stop or not at the end). Consequently, the user can generate drive cycles for a given route that has not been recorded in the training data, yet reproducing the naturalistic/stochastic nature of driving.

SVTRIP Target Trip Attributes
Fig.2 - Target Trip Attributes.

Development plan. SVTRIP is currently in development for public release. If you are interested in a demonstration or beta-testing, please contact Dominik Karbowski (Lead developer).

Technical Manager

Dominik Karbowski