|
|
|
|
           |
|
SoftThought, Inc. > Projects > Autonomous Vehicle "Solo" Autonomous Ground Vehicle Technical Approach We are currently using two temporary vehicles to perform software and hardware compatibility tests on. One is a full size All Terrain Vehicle (ATV) and the other is a vehicle on a small eighteen inch frame from a modified radio-controlled car. We emphasize that these vehicles are for testing only and that we plan to purchase and license a street-ready vehicle this fall for use in the Urban Challenge. Our research is based on providing systems which will work on a wide range of vehicles. Our systems can operate on either of these vehicles and will operate on our Urban Challenge vehicle. This research began eighteen months ago when we were entrants in the DARPA Grand Challenge 2005 as Team NOVA with vehicle SOLO. We have continued our work and have developed systems with the following safe baseline capabilities:
Our current system relies on visual interpretation of the environment. It has been successful in finding traversable pathways while following GPS checkpoints. It can maintain safe speeds and can stay on the roadway even with limited GPS. It has a preliminary collision avoidance system which forces the vehicle to make a hard stop or turn whenever an obstacle appears in the current pathway. Descriptions of each of the components in the architecture of Figure 1 follow along with references to our company statement of objectives. 
Figure 1. Upper level architecture. Hardware A Yamaha Rhino, named SOLO, which we have modified provides our main mobile platform. We also use a modified radio control car on which we strap our laptop computer and camera. Both vehicles can use the same computer and camera with our software. We have been able to use our prototypes to learn and share knowledge between the two vehicles. Our artificial intelligence interprets the environment and provides actions for the vehicle. When we change over to a full-size street vehicle this fall, we will add three more cameras, an electronic compass and a SICK laser range finder. The principle test vehicle we use is a Yamaha Rhino 660 All Terrain Vehicle which has been modified so that a computer can perform all normal driving functions (Figure 2). We are using a Galil Motion Controller to distribute commands to 4 AMC Power Amplifiers. These in turn supply power to the linear actuators and a rotary steering motor (Figure 3). The linear actuators provide changes in the gear shift, acceleration and braking while the steering motor provides the direction of turn. 
Figure 2. ATV autonomous vehicle. 
Figure 3. Vehicle actuators and servo motor. The electronics are contained in a secure box which is water resistant, cooled and easily accessed (Figure 4). It has safety shut off switches with individual fuse links. A generator supplies main power which is shunted through power supplies to normalize current for other components. It helps power our emergency flashing light and horn system used in autonomous mode. It will power our wireless e-stop to complement our manual e-stops on vehicle. 
Figure 4. Vehicle electronics bay. In addition, we have a small vehicle that can respond to the same commands. It is a platform from a small RC vehicle fitted with some or our electronics (Figure 5). Our equipment is connected to the vehicle servos and miniature actuators through a USB Hub. 
Figure 5. Small autonomous vehicle. The shocks were replaced with heavy duty springs to hold the heavier load of a laptop computer. We add our camera and laptop computer and the same software used with the large vehicle. It works in a smaller test area with its own roadway, goals and obstacles but can respond in the same manner that the large one does. Our point here is that we can transfer our system to other vehicles efficiently. Both demonstrate learning and intelligent responses. Research Collision Avoidance Collision avoidance is the most important aspect of this work. We are approaching this problem by identifying objects in our image ‘grid’. We are looking for objects in the video field and determining their locations image by image. In addition, we are designing to find rapid entry objects that might come from outside our main view. This is being accomplished by placing objects on roadway and varying location and size until we understand their impact. For the rapid entry items we will introduce obstacles from the side of the vehicle to the front in order to provide ‘surprise’. Safety Emergency manual E-stop pushbuttons have been installed on our test vehicle. These are red with clear labeling so that the will stop the vehicle immediately. When we acquire the new vehicle in the fall, we will purchase two more pushbuttons so that we can install a total of four on the vehicle. There will be one for front, back and sides wired so that any button can shut down the vehicle. Later in the project, we will add wireless control for the pushbuttons. Advanced Navigation and Traffic Current traffic tests are for static conditions. Advanced conditions include moving traffic. The range finder will be used at the front of the vehicle to augment the image processing of distances to objects. We will continue our techniques with objects moving in and out of our field of view and will extend that to all four sides of the vehicle. This will include special conditions at intersections where we will evaluate moving traffic and use that to find safe passage in our minimaps. This way we can safely deal with moving traffic conditions. We will test our ability to produce safe maps so that we can merge with traffic, cross it and safely proceed to a desired location whether in a parking lot or a large open area. Our test areas will be used in both static and dynamic conditions. We will continue these techniques where we will turn off GPS and navigate visually. We have demonstrated the ability to pull into parking areas and back out. This was accomplished earlier however, without traffic. Moving traffic will be introduced for advanced collision avoidance. The addition of cameras will allow us to see traffic all around the vehicle and improve safety. The cameras will aid for images at the side and behind the vehicle. This should enhance our ability to have safe behavior when passing, at intersections, and merging as well as parking. All conditions will be tested with the test vehicle first and then the newer one. The minimaps will be developed from views on all sides of the vehicle. The maps are linked together with heading and distance information. In this way, we can re-plan our travel paths whenever a roadway is blocked. We will perform these tests with and without GPS information. Copyright 2006 SoftThought, Inc. |
|