In a pilot project for collision detection, Autonomous Reply brings together special technologies such as computer vision, LIDAR and edge-to-cloud-processes via 5G, as well as deep learning for neural networks.
Together with the City of Regenburg and the University of Regensburg, Autonomous Reply is designing a safe, smart city with autonomous vehicles. The aim is collision avoidance, by involving all road users in the interaction between autonomously driving things. Previous danger zones, such as blind spots, lack of visibility on corners, or silent vehicles are thereby ruled out.
The idea in a nutshell: In addition to a sophisticated architecture within the autonomous vehicle itself, an app can warn vulnerable road users and prevent collisions with pedestrians, cyclists or scooters. Real-time processes are used to collect, process and send data providing well-timed information about possible collision courses.
In the closed test area, an industrial park at Regensburg, LIDAR sensors and cameras are placed on lamp posts to continuously collect data on the traffic situation. At multiple points, moving objects at an intersection are classified, for example, as pedestrians or cyclists.
Data from both types of sensors are sent to the cloud via an edge device. Neural networks use the data in predictive models to plan the route of the vehicle, thereby calculating all the dependent relationships in the entire system in real-time.
Computer vision and LIDAR sensors on the autonomous vehicle are used to monitor the traffic situation for the vehicles themselves, including position and speed.
The processed information is then delivered from the cloud to two types of recipients: it goes back to the vehicle and used to control it; or it can be transmitted via an app to all road users in the test area. In the event of possible collision courses, the app can issue a warning.
Artificial intelligence (AI) algorithms assemble and classify pixels from the camera and LIDAR data on the autonomous vehicle. Various objects, such as bicycles or pedestrians are thereby identified. Autonomous Reply assigns corresponding motion profiles to these objects.
8 TB of data per minute is sent via 5G to three edge points per traffic intersection. Autonomous Reply's NVIDIA-based edge system processes the data into object lists. These are then forwarded to the cloud.
The City of Regensburg retains data sovereignty. In compliance with GDPR, all the information collected is anonymised and stored in a smart city cloud. This is subsequently transferred to the so-called people mover cloud, which in the pilot project controls two autonomous vehicles.
Autonomous Reply uses deep learning, a special machine learning method, to train the neural networks. In the initial phase, this is done using synthetic data, for which Autonomous Reply programs simulations using MatLab and CarMaker. In the second phase, this simulation data is expanded by adding real data.
In the initial five-month phase Autonomous Reply has elaborated on the concept of the interaction between the complex subject areas, while finalising the simulations with the synthetic motion profiles for route prediction. The first tests using real data will be completed by the end of 2021. This will be followed by a one-year proof of concept.
8 TB
of data sent to the edge points per day
3 edge points
per intersection, to pre-process data
2 shuttles
controlled by the people mover cloud
3 years
project time
das Stadtwerk Regensburg GmbH is a wholly-owned subsidiary of the City of Regensburg. It is the managing shareholder of its wholly-owned subsidiaries. These include: Stadtwerk Regensburg.Mobilität GmbH, Stadtwerk Regensburg.Fahrzeuge und Technik GmbH, Stadtwerk Regensburg.Bäder und Arenen GmbH, and also includes Jahnstadion Regensburg and Stadtwerk Regensburg.Dienstleistungen GmbH.
Within the Reply Group, Autonomous Reply is the specialised company for the software and system integration of autonomous things. The experts advise companies in the industrial, automotive and new mobility sectors from the sensor to the infrastructure. The portfolio includes holistic solutions across the entire value chain – from strategy definition and advice on application possibilities to design and implementation. The offer includes edge computing, embedded software, cloud services and integration into different eco-systems. State-of-the-art technologies and methods from the fields of deep learning, machine learning and computer vision are used.