Application-specific Autonomous Systems
All From One Source
Pioneers of self-driving passenger cars repeatedly face setbacks. The media regularly report on tragic accidents. The control systems used do not yet seem to be fully up to the complex situations in road traffic. The situation is different in the industrial environment, where autonomous systems drive, fly or swim under controlled conditions. Here, the technology has long been established.
Picture 1: Autonomous Lawnmower Developed by Stettbacher Signal Processing AG.
Stettbacher Signal Processing AG (SSP) is a service provider in the field of sophisticated technologies, such as sensor and measurement technology, signal and information processing, control and drive technology, etc. All these technologies are used in autonomous systems. For this reason, SSP first received an order for the development of a self-driving rover several years ago. To date, SSP has successfully developed several application-specific vehicles for various customers.
It is usually the case that the customer already brings a detailed idea of the vehicle's mechanics to the first meeting, but is uncertain about the vehicle's control. But it is precisely this control that is the real challenge for autonomous driving, especially when the demands regarding driving accuracy and/or dynamics are high. This has always been the case in past projects, and based on this, it was decided in each case to design a dedicated and optimized control system. But let's take it one step at a time: When developing the control system for autonomous systems, a few fundamental questions arise first. The first is safety, for example, whether the vehicle poses a danger to people. If so, the safety aspects and, in particular, the applicable standards and approval procedures must be clarified first. The second main question is navigation. How does the vehicle recognize its position and how does it find its way to a destination? It is important to decide whether the vehicle will be operated indoors or outdoors (or both) and how large its radius of action will be. Furthermore, it must be decided whether the vehicle has to drive on predefined paths or whether it has to find its own way to a destination. In this context, it must also be decided which procedures will be used to prevent possible collisions.
So right at the start, numerous fundamental questions come up. When answering them, the costs must of course also be taken into account. And optional or nice-to-have features in particular must be weighed against their cost. In the end, technical design decisions are made, such as the type of power supply, the performance of the engines and the drive concept, etc. Of particular interest in autonomous vehicles is, of course, the method of pose estimation, which includes position and orientation. In this process, not only sensor data are read out and used, but the pose is optimally estimated mathematically from all available data (for example, from GPS/RTK, UWB, lidar, laser, gyro, odometry, etc.) in the so-called sensor fusion. This results in very accurate attitude data, even if the sensors measure inaccurately and are possibly still subject to drift and delay.
Another important topic concerns finding and approaching destinations. Often, a digital (land) map is used, in which any obstacles, driving paths and special locations, such as charging stations, parking and service areas, etc. are marked. Given a target coordinate, the robot either travels along a predefined path or calculates the path to the target using a path planning method. Typical methods for this are based on the Dijkstra algorithm. For each point on the map, a cost value is calculated that represents the smallest possible effort (often the distance, possibly taking into account other influencing factors such as gradients, curve radii, etc.) to get from the starting point to this point. In this way, one can find the optimal way to any reachable destination on the map.
Even if a digital map is available, unexpected obstacles can rarely be completely excluded. Therefore, in most cases a sensor system for collision avoidance is indispensable. In many cases a lidar is used for this purpose. As an inexpensive alternative, distance sensors based on ultrasound or mmWave radar can be used, and increasingly also cameras in combination with machine vision (AI).
If the path is fixed, the vehicle requires a driving controller that selects a suitable speed and steering strategy depending on curves, gradients, obstacles, etc. The driving controller is then used to control the vehicle. Various methods are known for this purpose, based on geometric, physical or statistical models, which are more or less suitable depending on the practical boundary conditions.
Stettbacher Signal Processing AG (SSP) has built up a large know-how in all these disciplines over the past years. Thanks to the experience gained from many different projects, the company is able to react quickly and purposefully to new challenges.
