|
刚才听写了一遍,有些地方听不清楚,可能涉及到陌生的术语。写下来的只作为参考,本人对笔误引发的任何误解不负责任,呵呵。
This video presents STARMAC, the Stanford Testbed of Autonomous Rotorcraft for Multi-Agent Control.
The testbed comprises six quad-rotor helicopters, two of which are shown here doing autonomous hover using GPS. Experimental applications for this
testbed range from decentralized collision avoidance, to autonomous search and rescue with a team of aircrafts, to studying the interactions between
humans and autonomous systems.
This aircraft design has evolved from many flight tests. The sensor sweep per every vehicle includes a GPS unit to sense position, an ultrasonic
ranger to sense the distance to the ground, and an inertial measurement unit to sense the acceleration and rotation of the vehicle. Additional
sensors are concurrently being carried in order to preceive the environment including cameras with both monocular and stereo vision. LIDAR range detectors, and avalanche rescue beacon receivers.
The vehicles are controlled in real time using a microprocessor to control the high rate dynamics, and using a high level computer to perform optimizations.
The quad-rotor is controlled by varying the relative speed of the motors. For instance, the rolling torque is controlled by decreasing speed of the left rotor and increasing the speed of the right rotor; Yaw is controlled by varying the relative speed of counter-rotating rotors; Vehicle position is controlled by varying the total of thrust and vehicle orientation. So that was the theory, but, only tried using this for the control system design, we found out that there were aerodynamic effects that we weren't accounting for. By modeling these aerodynamic effects, we designed controllers that reject these disturbances. ************** control system design uses feedback on angular acceleration and linear acceleration. As you can see, it's very capable of rejecting disturbances. The vehicle shown here is in autonomous hover using position data from a overhead camera system that does block tracking to replace GPS indoors. Using this system, multiple aircrafts are able to fly autonomously, safely and accurately in this confined environment.
Now of course sometimes, the opportunity for a more rigorous disturbance rejection testing arises. The vehicle's controlled here, by a human pilot issuing attitude angle command using a joystick.
The STARMAC quad-rotor helicopters are able to track ********* the varying attitude commands, as well as fly at reasonable speeds. With the root mean square error of less than 1 degree. This omnidirectional aircraft has the capability to rotate to face any direction while in flight. This feature makes it easier to use sensors for one of the fields of view. Here, a human pilot is actually controlling the attitude angles of the aircraft by issuing commands that rotate onto the body coordinate frame.
With the use of carrier phase differential GPS outdoors, the position can be maintained with approximately 50 centimeters of root mean square error as shown here.
We've performed optical avoidance experiments, with the vehicles required to fly around the boxes and not over them. During this test, the aircraft inadvertently wandered into a box, demonstrating the safety and robustness of the aircraft's design.
Here, an experiment using the tunnel Tunnel MILP is shown. The aircraft must navigate from the right side of the field to the left side of the field. To solve this optical avoidance problem, Tunnel MILP first finds the shortest path to the goal. Then it follows a tunnel aroud this path, and formulates a mixed integer linear program to fly the vehicle through this path under dynamic constraints, optimally.
Next, is a demonstration of decentralized collision avoidance problem. The three vehicles negotiate deconflicted trajectories using Nash Bargaining ****** , their initially conflicted paths across the field are resolved. In order to track other vehicles without using GPS, and in order to preceive the environment, we're beginning to use a camera system. The camera provides high resolution, high frame rate images directly to PC104 onboard the aircraft. One current goal for the testbed, is to actively seek information using the information theoretic contrl. This will involving pivoting the camera around and moving the entire aircraft, in order to gain a better perspective of whatever the vehicle is trying to learn about. We're currently working, ******automated search and rescue. Using avalanche rescue beacon transceivers, not humanned aircrafts. The vehicles communicate their observations with one another over a wireless, and track the target's location using a **********(Monicarold?) method called particle filtering. The vehicles are able to actively seek information using an information theoretic control lock, to quickly and accurately localize the target.
In conclusion, the STARMAC system, has provided a testbed with which can strive to push the limit of autonomy.
最后这一句也有点怪怪的,不通顺,他到底说什么呢? |
阿莫论坛20周年了!感谢大家的支持与爱护!!
知道什么是神吗?其实神本来也是人,只不过神做了人做不到的事情 所以才成了神。 (头文字D, 杜汶泽)
|