AD-Assurance-Lab/connected-eco-driving-doe-ee0008468

Connected eco-driving, Fort Collins 2019

Real-world, synchronized vehicle, ADAS, vehicle-to-vehicle (V2V) and infrastructure (V2I) data from urban drives in Fort Collins, Colorado, in 2019, collected for vehicle velocity prediction and predictive energy management. Part of the U.S. Department of Energy Vehicle Technologies Office project DE-EE0008468, "Mobility and Energy Improvements for Fort Collins and Denver" (Colorado State University with NREL, the cities of Fort Collins and Denver, and Western Michigan University). Described in:

Rabinowitz, Gaikwad, White, Bradley, Asher. Synchronous and Open, Real World, Vehicle, ADAS, and Infrastructure Data Streams for Automotive Machine Learning Algorithms Research. SAE Technical Paper 2020-01-0736, 2020. https://doi.org/10.4271/2020-01-0736

Code, documents and the full publication list: AD-Assurance-Lab/connected-eco-driving--doe-ee0008468--main.

How it was collected

An instrumented Chevrolet Malibu (the ego vehicle) drove loops on arterials around central Fort Collins (College, Prospect, Shields, Mulberry, Drake, Lemay) with a Mobileye 6 camera, Bosch mid-range radars, a ZED stereo camera and a CAN/OBD logger. Volunteers drove their own cars nearby with Freematics GPS/IMU/OBD loggers to stand in for V2V messages. The City of Fort Collins supplied traffic-signal phase and timing and BlueTOAD segment travel times.

Day	Folder	Ego	V2V drivers	Infrastructure
2019-05-15	raw/2019-05-15/	ADAS log, ZED video	D01-D04	signal timing, segment speeds
2019-07-03	raw/2019-07-03/	logger, ADAS log	D02, D03, D05, D06	signal timing, BlueTOAD segments
2019-08-01	raw/2019-08-01/	logger, ADAS log	D02, D03	signal timing, segment speeds
2019-10-21/22	raw/2019-10-21_22/	u-blox GNSS (vehicle signals in processed/ and the table)	aggregated only	SPaT logs 13:00-19:00, segment speeds

The ML-ready tables (speed-prediction/)

One row per sample along each trip, sampled by distance travelled; trip numbers the laps. Columns: the ego vehicle's position, speed and engine speed, plus the accelerator pedal in October and engine load and throttle in August (all readable over OBD-II), then signal_k_* for the next five signals along the route (index, distance to it, current phase), segment_* (current segment, its measured speed, distance remaining), v2v_mean_speed and v2v_sigma_speed (statistics of the volunteer traces at this location) and, in August only, lead_* (lead vehicle offset and speed from the aftermarket ADAS sensors). Missing values are NaN. Units follow the source: SI unless stated, segment speeds in mph, time in seconds from midnight. Phase numbers follow Fort Collins' standard (1 NB left, 2 SB through, 3 WB left, 4 EB through, 5 SB left, 6 NB through, 7 EB left, 8 WB through).

from datasets import load_dataset
ds = load_dataset("AD-Assurance-Lab/connected-eco-driving-doe-ee0008468", "speed-prediction")

Everything else

• raw/: the logs as recorded. ADAS logs are Vector .blf with MATLAB/CSV decodes; logger CSVs from the volunteer cars use Freematics' format (time, PID, value; PID A latitude, B longitude).
• processed/: the team's synchronized and interpolated MATLAB structures per day, and processed/nn-training/ with the cleaned traces used to train the first networks.
• results/: velocity-prediction results (October data).
• video/2019-05-15_zed-stereo.svo: ZED stereo recording; open with the ZED SDK.
• maps/: the route with signals and segments.

Privacy

Volunteer drivers appear only as D01 to D07, in file names and inside the MATLAB data. Every GPS fix in their traces lies inside the study area; none starts or ends at a home. The driver-to-car list is not released.

Not included

Argonne test data, TomTom map data (licensed), vendor documents, project management files, and the ego vehicle's own bus data beyond the standard OBD-II set. The Malibu was a GM-supplied EcoCAR vehicle, so of its signals the release keeps only what any OBD-II reader gives (vehicle speed, engine speed, accelerator pedal) and the position fix; its CAN logs, other channels and radar tracks are not released, and the network input matrices were removed from results/.

License and citation

CC BY 4.0. Copyright 2019-2021 Colorado State University and Western Michigan University. Please cite SAE 2020-01-0736 above. This material is based upon work supported by the U.S. Department of Energy's Office of Energy Efficiency and Renewable Energy (EERE) under the Vehicle Technologies Office award number DE-EE0008468.