Robotic Systems ECE 401RB Fall 2007

Transcription

1 The following notes are from: Robotic Systems ECE 401RB Fall 2007 Lecture 14: Cooperation among Multiple Robots Part 2 Chapter 12, George A. Bekey, Autonomous Robots: From Biological Inspiration to Implementation and Control, The MIT Press, I. Communication Among Multiple Robots This is a complex issue. Communication is essential between robots if they are to accomplish some global goal. The form that the communication should take is not clear. Three possible methods of communication. 1. Point-to-point communication Individual robots communication between each other. Passing information such as goal locations and locations of obstacles. 2. Broadcast Sending messages to all members of a group. 3. Communication via the environment Implicit communication Implicit = Implied or understood though not directly expressed Ants leave chemical trails. Animals frequently mark boundaries with urine or other chemical markers. Lecture 14, Page 1 of 11

2 Sometimes direct communication is not necessary The high-level goals of the group are established in the rules of behavior. These rules are applied by the group members and they end up achieving the goals. No explicit communication among the members is necessary. African termites build large mounds in this way. Each termite operates on its own rules. Carry the sand as high as possible on the mound, then drop it. Humans frequently infer the emotional state of another person without intentional transmission on the part of the sender. Examples are facial changes due to embarrassment or perspiration due to fear. Internal states of an animal might be inferred from how it is walking. Due to age or injury, it might be walking slowly. Making it a more likely target for predators. Cooperation without Communication Consider accomplishment of a task by a colony of robots without explicit communication. Arkin has shown that the task can be accomplished this way. Each robot simply acts upon its own perception of the world. A form of cooperation emerges. Robots converge to work on the task. Balch and Arkin studied a more complex situation. Foraging plus consuming and grazing. Attractors are items of interest for the foraging or food for the consuming task. - The number of attractors and their mass are configurable parameters. Grazing involves finding an area that has not been completely covered with tracks. Robots wander until they are able to begin one of the three tasks. Lecture 14, Page 2 of 11

3 Three forms of communication are studied. - No communication - Each robot uses its own sensors to find other robots, obstacles, or attractors, with no help from other robots. - State communication - Robots communicate their internal state whether they are in a foraging, grazing, or wandering mode. - This can be explicitly communicated through some kind of message. - Or implicit by observing how the robot is behaving. - Goal communication - Information concerning goals is transmitted intentionally. - This can be goal locations, type of goal, presence of predators, etc. - Bees use explicit goal communication by executing a dance at the entrance to the hive. - Parameters of the dance provide direction and distance information to the source of nectar that was discovered. Lecture 14, Page 3 of 11

4 - Results are given below. - Communication improves performance significantly in tasks with little implicit communication, such as foraging and consuming. - Communication appears unnecessary in tasks for which implicit communication exists such as grazing where robots can see where others have been. - More complex communication strategies, such as goal communication, may offer little or no benefit. Other studies have also shown how performance for a group of robots is enhanced with communication. Lecture 14, Page 4 of 11

5 II. Formation Control Organized patterns of movement occur frequently in the animal and human world. One example is birds flying in a V formation. What other examples are there? Military movements. Aircraft formations. Animals surrounding their prey. Sports. Lecture 14, Page 5 of 11

6 Why are these formations helpful? Share leadership roles. Provide a stronger team. Protect weaker members. Heart rates drop 30% in birds flying in formation. Formation Control Using only Local Information Robots can create formations based on simple rules. Unit center referenced Each robot decides its position relative to the centroid of the group of robots. Leader referenced Each robot uses the leader s known position Neighbor referenced A neighbor s position is used as a reference for the robot s own position. Each robot can keep a friend at a desired angle and distance. Lecture 14, Page 6 of 11

7 Several formations are possible. These are based on simple algorithms that do not require global information. Global Approaches to Formation Control More generally, even if global information is available like GPS, there are three fundamental approaches. Leader following Followers track the position and orientation of the leader. This is very simple. But leaders do not get feedback from followers. - So they might move too fast and lose followers. It is also a problem if a leader fails. Lecture 14, Page 7 of 11

8 Behavior based architectures Several behaviors are prescribed for each robot in formation. - Such as obstacle avoidance, goal seeking, and formation keeping. The selected behavior is a weighted combination of possible behaviors. However, the group behavior that results from this approach can be hard to specify mathematically. Virtual structure based The entire formation is treated as an entity. - Then it is easy to define how the entity as a whole should move. Motion and control of individual robots is derived from the motion of the entity. Virtual structures must be dynamically reconfigured as robots are added or removed. III. Robot Soccer Use of multiple autonomous robots in teams to play to a competitive game such as soccer is a major challenge. Real time Highly dynamic environment Need to have closely integrated perception, reasoning, and action. While maintaining communication with one another. This assumes soccer players should actually work together! Must adapt strategies to the actions of the opposing team. Lecture 14, Page 8 of 11

9 There have been international RoboCup competitions. First generation of robot soccer players. Used a camera over the playing field. Tracking positions of ten robots and the ball was difficult. - Each robot needed an identifying marker. After image processing, then commands needed to be given to each robot. More recent years On-board capabilities of robots have increased. - Both perception and processing power. Increasingly autonomous Less reliant on remote commands. Use color landmarks on the field to determine their position and actions. Sony AIBO robots have been used in legged robot competitions. - The robots are fully autonomous. - All robots have the same capabilities. - So the difference lies in cognitive ability, strategy, and teamwork. Lecture 14, Page 9 of 11

10 Future Humanoid robots like ASIMO? Full-sized humanoid robots competing with humans? MUCH progress is needed before that happens. IV. Other Issues Task Assignment Teams can be heterogeneous or homogeneous. If heterogeneous, one might be a scout, another carries supplies, and another could be a beacon to transmit goal information to robots on the team. Designers can assign tasks to these robots. But a more innovative approach would be to have to robots decide among themselves. Then there must be a way for robots to communicate among each other about their capabilities. Then they can ask for help. - Or they can negotiate on which tasks they wish to perform. - Maybe through a bidding process. A possibly more complex issue is how to give robots tasks they do not wish to perform. - Or choose a second-best robot to perform a task if the preferred robot is too busy. This is a largely unexplored area of research for robots. A higher level issue is also how to explain to a group of robots the task they are to accomplish. Reliability and self-repair One of the main reasons for multiple robots is higher reliability. If one fails, others can take over. Research issues How robots take over tasks for other failed robots. How to find a robot capable enough to take over. If a robot needs to be reprogrammed to take over the task. Also, there may be a need to remove damaged robots so they do not get in the way of other robots completing their tasks. Lecture 14, Page 10 of 11

11 Mutual recognition How do robots recognize other robots? Whether they are friend or foe, team member or the competition. If implicit communication is to be used, how does one robot watch another to understand its actions? To simplify, how can the actions of other robots be classified into a small set of options? Like whether another robot is foraging, grazing, or consuming. Localization The ability of a robot to determine its location in space is essential for navigation, goal seeking, and other activities. Localization can be absolute. Using GPS, for example. But it could also be relative. With respect to landmarks or other robots. Relative localization is more challenging if the other robots are moving. Ongoing research challenges Collaborative behavior in robots will continue to be a major challenge for the next decade or two. Particularly for cases where humans and robots are to collaborate. Lecture 14, Page 11 of 11