The solution is compatible with different models including Lincoln, Geely, Ford, and Dongfeng.
DeepRoute.ai’s solution can safely maneuver on highways, in heavy rainfalls, at night and in other inclement weather conditions.
In April 2021, it took 3 days to deploy the fleet in the CBD area of a new city, Hangzhou.
Multi-Sensor Fusion Technology
DeepRoute.ai’s sensor fusion technology can detect various objects over 200 meters away accurately and efficiently to guarantee safe driving, even in inclement weather, at night or in other scenarios.
The deep learning algorithm has ranked No.1 on global leading autonomous driving dataset KITTI and Semantic KITTI, and the papers were accepted respectively by CVPR 2020, CVPR 2021 and ICCV 2021, one of the most important conferences worldwide in the field of computer vision.
DeepRoute-Engine improves the efficiency of inference and allows the perception algorithms to run efficiently and smoothly on a low-cost, low-consumption computing platform. Its speed is six times faster than existing inference engines in the mainstream deep learning framework.
It is also compatible with various computing platforms, thus reducing dependency on high performance chips caused by short supply.
Human-Like Planning and Control
Based on game theory, DeepRoute.ai planning and control algorithms can think like a human to make real-time decisions in complicated road situations. Traffic dynamics, or the interaction between self-driving cars and other road participants, are perceived to make safe, efficient and smooth driving decisions. The algorithm also plans real-time optimal trajectory for the vehicle, overtaking or avoiding cars in certain situations. It balances driving efficiency and comfort while prioritizing safety.
DeepRoute-Tite is a power-efficient computing platform that serves as the brain of the self-driving system. It operates on less than 150-watts and operates at only half the cost of existing solutions. The compact design can cool off quickly and allows more room for storage.
DeepRoute-Syntric not only synchronizes and distributes time and space sensor data, but also monitors the data in real time, and controls the vehicle when malfunctions or unsafe conditions are detected. Syntric supports different models and brands of sensors. DeepRoute.ai can quickly adjust sensor models and layout based on project cost, car model and other needs.
DeepRoute-Vision is a vehicle camera with a high dynamic range of 120dB. It accurately captures the surrounding objects even in strong sunlight. Its automatic image adjustment and anti-flicker functions ensure an accurate capture of traffic signs. DeepRoute-Vision supports image compression of different formats, saving bandwidth and transmitting multiple camera data to the computing platform in real-time.
Taking advantage of the low-latency and high bandwidth of 5G, the cloud monitoring system can transfer operational data to the control center in real-time for staff supervision. In the future when no safety driver is in the vehicle, the vehicle can be switched to remote mode to ensure safety in the event of malfunctions or long-tail scenarios. Additionally, in the case of a public security incident, DeepRoute.ai can utilize cloud monitoring to control vehicles to provide transportation services for patients with contagious disease.
The main system and backup system operate independently to provide a multiple safety guarantee.
Sensor data from 5 LiDARs and 8 cameras are fused and combined with an HD map, allowing complex road data to be perceived and efficiently inferred. Dynamic objects are detected and avoided automatically.
Shenzhen has set up a cloud monitor system to supervise test and operational vehicles in real time, allowing for intervention when needed.
The safety operator will take over immediately in the event of any failures or emergency situations.
Records driving data for safety improvements and protects user privacy.
Defends against cyber attacks to guarantee safety of the system, data and telecommunications.