|
Small Scale Simulations
We have simulated in great detail the control of low numbers of vehicles (up to 20) navigating throughout a building. A Sandia-developed simulation environment called Umbra is used to test the convergent control algorithms in a complex environment. Umbra allows the simulation of multiple autonomous agents with a variety of physical phenomena such as RF communications, interactions with solid objects (i.e. collisions), ultrasound ranging, IR detection of objects, vehicle physics, terrain descriptions, and other phenomena the user wishes to study. All of these physical attributes can be simulated simultaneously with a graphical visualization that allows the monitoring of the vehicles' performance over the terrain.
Such a simulation was implemented for the case of multiple, small, wheeled vehicles traversing a single floor in a building with multiple corridors, rooms, and entrances. The vehicles are modeled after the vehicles that will be used in the hardware tests. Each vehicle contains 4 IR sensors for detecting objects between 6" and 2.5' on all 4 sides of itself. The vehicles also contain RF communication devices to be able to converse with other vehicles within a 100' line of sight range or roughly 30' through walls. They also have ultrasound capability to measure the distance between them provided they are within 30' of each other and in line of sight range. The vehicle physics are quite simple and proved adequate on a smooth surface. The building model was generated as a CAD model and contains several connected hallways as well as a multitude of variable size rooms. The control algorithms for the vehicles must avoid contact with walls and other vehicles. Beyond that, the control goals can vary depending on the motives of the operator. For instance, the vehicles can spread out to provide maximum coverage of the building, or they can stay within a prescribed area, or they can maintain a particular formation.
The restriction that vehicles can't run into walls, doors, or each other essential ensures they remain inside the building. This is accomplished via rules that use the IR sensors to follow walls down a hallway. This will enable the vehicles to move throughout the building, though not necessarily in any prescribed fashion. Further restrictions on the vehicles involve the maintenance of a continuous RF communication network. This requires that vehicles stay within 100' of each other or less if line of sight (LOS) is lost (i.e. they may have to stay at a wall junction to maintain LOS). A more stringent condition is the ability for each vehicle to know its absolute (x,y) position with respect to some global coordinate system. This requires triangulation off of two known vehicles using ultrasound as a distance measurement. This implies that at least two vehicles must remain in fixed known locations until the other vehicles can triangulate off of them. There are a number of techniques to accomplish this that were investigated in Umbra. These include the law of cosines triangulation, steepest descent triangulation, and conjugate gradient triangulation. All had advantages and disadvantages depending on the number of vehicles and the on-board processing power and memory. Finally, there is the constraint that the vehicles spread out and "cover" the building uniformly. A gradient-based scheme was used to repel the vehicles from each other to diffuse through the building while the aforementioned constraints keep them close enough to communication and compute absolute position.
Figure 5. Detailed simulation of multiple vehicles navigating a building. The protruding green and blue cones represent the 4 IR proximity sensors. |
