Abstract. Since research in intelligent autonomous road vehicles gets more and more popular, many interested researchers are confronted with the needs to setup a commercially available vehicle with drive-by-wire. Up-to-date road vehicles contain many mechatronical components, sensors and driver assistance systems, which can be interfaced and thus reduce the needs of expensive modifications and additional actors to a minimum. This paper describes how to interface steering, throttle and brake as well as built-in sensors, and shows an inexpensive way to provide a safe research platform for road traffic.
Keywords. Drive-by-Wire, Autonomous Vehicle, Vehicle Control, Driver Assistance, ABS, ESP, Electric Power Steering
1. Introduction
In the last decades, vehicles have been equipped with more and more mechatronical components. Starting with conventional cruise-control over first safety systems like Anti-Blocking-System (ABS) or Electronic Stabilization Program (ESP) towards higher-level assistance systems such as adaptive cruise control (ACC), lane change warning systems or semi-autonomous parking assistants. Development of such intelligent assistants also rose interest of various research labs and universities, and established a connection between robotics area and automotive industry. Recent events like the Grand Challenge show this trend of applying the experiences gained from mobile robot research to road vehicles. It can be expected, that traditional robotics research topics like computer vision, object recognition, situation estimation, decision control as well as knowledge extension and learning will be included in future driver assistance systems. With integration of cognitive capabilities, such systems will be able to understand and evaluate traffic situations and derive reasonable and safe behaviors, which will be the basis for fully autonomous road vehicles.
Where automotive industry sets their focus of research on applying existing technologies to new driver assistance systems, universities and research centers are able to deal with more
Figure 1. Experimental Vehicle
complex problems thatdo not need to result in products in the near future, which means 4-5 years. It will be their major task to perform basic research in the field of cognitive road vehicles. This includes finding answers to questions of how traffic situations and vehicle knowledge can be represented, which learning algorithms qualify for these systems, and, how safety of autonomous vehicles can be verified and numeralized.
To apply the experience gained in the fields of mobile and humanoid robotics to road vehicles, the Institute of Computer Science and Engineering [1], and the group Interactive Diagnosis- and Servicesystems [2] started a collaborative project, the Smart Roadster Project, to extend their research towards this area.
A Smart Roadster (see Figure 1), which was donated by the Smart AG, will serve as experimental vehicle. Aim of this project is to setup a fully autonomous road vehicle, and to equip it with cognitive abilities to negotiate in road traffic environment. While autonomous lane-following or lane-changing maneuvers on highways were performed by various research groups [3,4,5], the emphasis of this project is to develop a cognitive platform with understanding for a wide variety of traffic scenarios, including inner-city and interurban driving. Even if this goal seems to be very far, and it cannot be expected to see autonomous vehicles in traffic within the next two decades, it is clear that the trend goes towards this direction. Therefore, research in this area is necessary and will be very helpful to improve assistance systems on the way to a fully autonomous vehicle. Figure 2 shows the architecture which is going to be used in Smart Roadster project. The interaction with the physical world occurs through a Sensor/Actuator Level. A camera-head with stereo and tele-cameras is going to be used as main sensor. Even though other sensor concepts like Lidar or Radar devices might be added in the beginning, the authors think that in the long term non-invasive sensors need to be preferred.
At the next level, sensor data is processed to obtain information about infrastructure like signs, street course and other static obstacles, as well as pedestrians, cyclists, other cars and moving objects. The control part in Figure 2 generates control values for throttle, brake and steering according to higher-level input, taking into consideration the vehicle dynamics.
The third level from the bottom contains the database and behavior execution. The database contains spatiotemporal information about recognized objects, their state and behavior as well as the own vehicle state. All capabilities of the vehicle are stored in a behavior execution net. It has an internal structure that consists of different layers of behaviors, which have a pre-defined coupling and progression as they are active.
The cognitive level contains the most important components: The scene estimation to understand and to evaluate a traffic scenario, and the behavior decision to trigger an appropriate vehicle movement. It is obvious that all possible situations can neither be identified nor stored. Also, there are many different ways of executing behaviors and reacting in situations. Therefore, the questions of how to learn unknown situations, how to improve the behavior, and how to assert safety for this extended knowledge will be the key task in this level.
According to the architecture described above, the development of our cognitive car can be devided into the following phases:
1. Phase: Setting up drive-by-wire capabilities
2. Phase: Implementation of sensor data processing and vehicle control to provide basic features
3. Phase: Integration of simple situation estimation to perform low-level driving behaviors
4. Phase: Extension of situation estimation to wide area of traffic situations and derivation of complex tasks
5. Phase: Evaluation and integration of safe learning algorithms to improve cognitive abilities
In the first phase, the goal is to provide a complete drive-by-wire ability for the system, and to make necessary modifications to guarantee safe operation.
Where in former times researchers had to make expensive modifications to setup a fully controllable road vehicle, up-to-date cars have included many assistance systems and mechatronical components. These systems include various sensors and actuators, and provide interface options which reduce modifications to a minimum.
Joachim Schr¨oder, Udo M¨uller, R¨udiger Dillmann
Institute of Computer Science and Engineering.
University of Karlsruhe (TH). Karlsruhe, Germany.
Tel.: +49 721 608 4243; E-mail: schroede@ira.uka.de.
No comments:
Post a Comment