EXERCISES
3.1 CONTROL PARADIGMS
A
mobile robot is designed to accomplish various tasks. The easiest ones involve moving around while avoiding obstacles and reaching a given goal point. The more involved tasks may include following a trail of an odour, mapping an unknown terrain, responding to voice and gestures, attending to man-centric environments as museums, theaters or airports etc. The simplest robot is at least a sensor actuator pairing. More sensors and actuators enhance a robot’s performance and add to its versatility. Other important parameters are the constraints of the robot and the constraints of the process, safety concerns, self preservation and optimised performance. This chapter focuses on control paradigms and architectures and used in robots and tries to implement the concepts of the last chapter in the engineering domain.Three important aspects of a good control paradigm are closed loop control, real time response and robustness to overcome sensor errors. Closed loop control is critical. This can be seen with an example: one of the simplest robot is the line following robot. The robot is fitted with a light sensor and it tracks a line marked in a certain colour usually black or white. Any undesired deviation away from the line is prevented by the sensorimotor pairing between the light sensor and the motor. An ANIMAT corollary would be with a hypothetical animal which has its eyes set on the road, as shown in Figure 3.1. The contrast in information flow for closed loop and open loop is shown inFigure 3.2. Without the light sensor, the robot will fail to track the line and it will move arbitrarily in the direction the heading is pointing to. Lack of a feedback mechanism1 will not allow the
1The concept of feedback is of importance in control systems and electronics, and closed loop systems which work on feedback are demonstrated to be more stable than open loop systems.
87
FIGURE 3.1 ANIMAT approach to line follower, an artificial animal keeping an eye on the road. This model is equivalent to a basic line follower with one light sensor. However, such a design with only one light sensor will not be of much practical utility. A robust design will have a ring of light or ultrasound sensors and also other sleek hardware, backtracking means and sensor fusion which will incorporate for the effects of friction, uneven terrain and other environmental factors.
robot to interact with the environment, explained in flowchart. Mobile robots are expected to have a real-time response with minimal lag. The sensor actuator pairing is modulated by a microcontroller and structured with a set of control rules and parameters. Often, it is a tradeoff between accuracy and lag time. A third concern is, sensors are liable to be faulty and give erroneous and noisy readings. Under such circumstances the system needs to continue to function well within a domain practicality. Therefore, mobile robot control should be organised with a parallelism approach which incorporates graceful degradation and elegantly handles exceptions so that failure of a single hardware or software module doesn’t hinder the broad goals of the robot.
To design a robot, there are three broad approaches. The first one is called sense-plan-act and it follows the concepts of traditional AI. It was developed in the early 1950s and processing sensory information to yield plans and maps was the basis to this approach. The robot has to develop exhaustive, near approximate maps of the world it has to navigate, where buildings are approximated as cuboid, clouds as spheres, a person is an oval head with a cylindrical torso and smaller cylindrical arms and roads as an array of flat straight sections etc. Once such maps have been made, the robot must find the suitable path to the goal point.
This approach can be summed up as a Minecraft world married to Google Maps. Even the best computers in the late 1960s could not have made maps like the current Minecraft games or Google Maps, but the idea was very much in the same lines. As discussed in the first chapter, Nilsson [253] designed Shakey’s navigation by making (a priori) grid-based maps.
The method relies on seeping in the knowledge from the environment and then processing that information to move around intelligently. Since space is apparently infinite compared to a sensor’s range, this would be implemented as a cyclical process where the sensors sense the environment, the microprocessor develops plans to navigate in this environment and then finally the actuator would trigger the motion.
The second approach traces the principles of behaviourism [52, 53], where instead of making a priori maps, the robot starts moving around on the merit of the sensory-motor
89
FIGURE 3.2 Closed Loop.In (a) the light sensor allows for interaction with the environment, the line follower tracks the line and follows it to the end; while without a light sensor in (b) the robot fails to have any knowledge of the environment and randomly moves without any goal or motive.
The stop for such a robot will not be graceful, probably ending in a collision.
pairing in the local environment with near real time sensor reading paired with immediate actuation and no appreciable plans or maps, akin to animals interacting with their local environment. This approach eliminates planning, at least a priori and deals with the dynamism of the environment in real time as an immediacy. This approach is very effective in navigation and other lower level tasks, but as we will see, it cannot be extended to higher level tasks.
A third paradigm, has aspects of planning as well as behaviourism. Such architectures are seen in most of the state-of-the-art robots. It will be discussed in later chapters that hybrid architectures are of pre-eminence for human-centric robots and robotethics.
1. Deliberative approach, detailed planning prior to execution. When implemented in a navigational context the robot has to design elaborate maps of the environment before starting to move.
2. Reactive approach, the agent acts per the merit of the situation, no appreciable planning is involved. Draws motivation from the animal and insect world, see obstacle - turn leftis probably the simplest reactive paradigm. When implemented across various sensors, over a variety of parallel tasks, a reactive paradigm leads to thebehaviour-based approach.
3. Hybrid approach, As a combination of the two, deliberative planning is executed in a reactive manner.
The next section is about a number of thought experiments by neuroscientist and cyberneticist, Valentino Braitenberg and serves as a wonderful illustration for development
The complete control mechanism for the agent can be designed by stacking together such SR diagrams as is briefly done for Braitenberg vehicle-3c in Figure 3.6. Traditionally simple behaviours are put at the lower ends with sophisticated behaviours at the upper ends, as shown inFigure 3.19.
An equivalent algebraic approach to behaviour is using functional notation;
β(s) =r (3.1)
Where behaviourβ with given stimulussyields responser
of behaviours from simple sensory motor design. It extends the description of the reactive approach to designing robots.