RoboSense Admits Long Way to Go For Driverless Tech
Sensors used by autonomous vehicles’ LiDAR and artificial intelligence systems still need serious development before driverless journeys can become commonplace.
Leilei Shinohara, vice-president of LiDAR maker RoboSense, admitted that in an exclusive interview with TU-Automotive. He discussed the key areas in which LiDAR sensors need to improve and said their “intelligent perception algorithms still need further testing to evaluate and improve performance.”
Deep learning
He also said AI is crucial to present and future AV development and AVs’ AI systems would need to use deep learning algorithms more in future.
“AI for environmental perception is very important. It is already doing a good job. However, due to complex environments, it is very challenging for AV sensors to perceive and understand the surrounding environment. Therefore, large data-driven deep learning algorithms based on a huge amount of road are needed. Most AI perception algorithms are video-based.”
Price challenges
Shinohara claimed LiDAR sensors offer AV developers better object recognition than other sensors but LiDAR makers’ “biggest challenge” is making the technology cheaper.
“Conventional sensors, such as radar, camera, and ultrasonic, all have their limitations. Cameras have difficulty detecting during low-light conditions. Radar has difficulty detecting non-metal and non-moving objects. When both conditions are met, AVs cannot detect anything. LiDAR is the best sensor to compensate for other sensors’ weaknesses. However, currently the biggest challenge for LiDAR companies is to reduce the price to be more affordable for standard passenger cars.”
