Research Focuses on Safety of Electric Scooter Riders, Other Vulnerable Road Users

Toyota and collaborators are working to advance safety for all road users with information from a research project.

June 28, 2024

Kami Buchholz

In this trajectory estimation demo, real world images are being used to form the basis of trajectory estimation for electric scooter riders and other vulnerable road users. (Purdue University)

Passenger vehicles traveling alongside electric scooters is an increasingly common road sharing scenario rife with safety issues such as an e-scooter rider suddenly veering into vehicle traffic. But what if an e-scooter’s sudden movements could be predicted before a collision with a vehicle? That’s the goal of ongoing research.

Toyota Safety Sense . . . A future generation of Toyota Safety Sense, an advanced active safety package, will benefit from CSRC research projects. (Toyota)

Toyota and collaborators from Purdue University recently collected real-world data relating to e-scooter riders, pedal bicyclists, and other vulnerable road users (VRUs) via a GPS and LiDAR-equipped vehicle fitted with six high-speed, high-definition cameras on the roof. The research vehicle captured more than 8,000 e-scooter images across the United States, including from cities in southern California, the Northeast, and the Midwest. The 200-plus hours of data collection resulted in 31 terabytes of visual and location data.

“This is the first and only large scale e-scooter detection dataset that is publicly available,” said Rini Sherony, senior principal engineer for Toyota’s Collaborative Safety Research Center (CSRC). The 8,000 e-scooter images and 8,000 non-e-scooter images are a start-point for training an artificial intelligence (AI) model to create VRU trajectory prediction algorithms. It’s relevant to have images with and without an e-scooter for AI model training. “Otherwise, the accuracy is not good,” Sherony said.

Sherony and others spoke with SAE Media during a June press event at the Toyota Motor North America (TMNA), Research and Development center in Ann Arbor, Michigan.

According to Dr. Yaobin Chen, professor of electrical and computer engineering at Purdue University and director of the Transportation and Autonomous Systems Institute, the ongoing CSRC research project involves identifying various VRU scenarios to help develop next-generation safety systems. “You need standardized scenarios and testing protocols in order to evaluate [safety system] performance,” Chen said.

VRU path projection of up to 1.5 m/4.9 ft accuracy is possible with video-only data that’s 2 to 4 seconds on the horizon. The next stage of the project would provide an up to 7-second trajectory prediction timeline with added sensor inputs from radar, LiDAR, and high-definition maps.

“The earlier we can predict [the trajectory of] vulnerable road users then the vehicle’s safety system can take action to prevent or mitigate a crash,” said Sherony.

Project researchers will also take more than 500 micro Doppler radar measurements of 30 different VRUs at different distances and motion directions. The micro Doppler signatures could improve VRU detection accuracy. “When you bike, the leg motion generates a very specific pattern from the radar,” Sherony explained.

The VRU study is one of 15 new CSRC collaborative projects aimed at advancing automotive safety industrywide. According to Jeff Makarewicz, group vice president of technical resources at TMNA, R&D, “We want to amplify the human’s ability to drive and be safe in a vehicle, and we want to protect all road users, including bicyclists, e-scooter riders, and pedestrians,” Makarewicz said.