A vision-based formation control framework
Summary (3 min read)
Introduction
- These controllers can be either centralized or decentralized and are derived from input–output linearization [10].
- Before describing the individual components of their control framework, the authors list several important assumptions concerning the group of robots and the formation.
A. Basic Leader-Following Control
- Note that this relative bearing describes the heading direction of the follower with respect to the leader.
- A complete stability analysis requires the study of the internal dynamics of the robot, i.e., the relative orientation .
- Assume that the lead vehicle’s linear velocity along the path is lower bounded, i.e., , its angular velocity is bounded, i.e., , and the initial relative heading is bounded away from , i.e., , for some, also known as Theorem 1.
- If the control input (3) is applied to , then the system described by (2) is stable and the outputin (4) converges exponentially to the desired value.
- The authors can study some particular formations of practical interest.
B. Leader-Obstacle Control
- This controller (denoted ) allows the follower to avoid obstacles while following a leader with a desired separation.
- For this case, the kinematic equations are given by (8) where is the system output, is the input for , and , .
- This occurs when vectors and are collinear, which should never happen in practice.
- By using this controller, a follower robot will avoid the nearest obstacle within its field of view while keeping a desired distance from the leader, also known as Remark 4.
- This is a reasonable assumption for many outdoor environments of practical interest.
C. Three-Robot Shape Control
- Consider a formation of three nonholonomic robots labeled , , and (see Fig. 2).
- Another approach that is more robust to noise is to use a three-robot formation shape controller (denoted ), that has robot follow both and with desired separations and , respectively, while follows with .
- As before, the authors will show that the closed-loop system is stable and the robots navigate keeping formation.
- Hence, it is preferred when the separations between robots are small, and when, coincidentally, the estimates of distance through vision are better.
- As in any practical system, unmodeled dynamics and measurement errors will degrade performance.
D. Extension to Robots
- Results similar toTheorems 1and2 are possible for formations of robots, but they have to be hand crafted, i.e., there currently are no general results.
- This, in turn, means that and will have to be appropriately constrained, e.g., and .
- For the same leader trajectory, notice the higher transient formation shape errors for the control graph (a). controllers depends on the length of the path for flow of control information (feedforward terms) from the leader to any follower in the assigned formation.
- In Section II, the authors have shown that under certain assumptions a group of robots can navigate maintaining a stable formation.
- The authors first illustrate this approach using three nonholonomic mobile robots , , and .
A. Choice of Formations: A Switching Strategy
- The authors consider the problem of selecting the controller, for robot , assuming that the controllers for robots have been specified.
- This intuitive procedure may fail if the switching strategy is not properly defined.
- Moreover, if the assumptions inTheorem 2are satisfied for subsystem , then .
- Since is common for all modes, the authors only need to consider in (14) for studying the stability of the switched system.
- (b) Fig. 8. (a) Six robots have an initial configuration close to the desired formation shape (an equilateral triangle with equally spaced robots).
B. Formation Control Graphs
- When , the authors can construct more complex formations by using the same set of controllers and similar switching strategies.
- Fig. 7 shows a directed graph represented by its adjacency matrix (see [19] for definition).
- For a formation of robots, the authors can consider a triangulation approach and Fig. 5 can be used to assign control graphs for labeled robots.
- Sample ground-truth data for trajectories for a triangular formation.
- From the omnidirectional imagery acquired by these cameras, the authors have developed several logical sensors—an obstacle detector, a collision detector, a decentralized state observer, and a centralized state observer (see [29]).
A. Decentralized State Observation
- The controllers described in Section II require reliable estimates of the leader robot’s (’s) linear velocity and angular velocity by the follower robot and their relative orientation ( ).
- Using the kinematic (1), their extended state vector then becomes (17) (18) where , is the process noise, is the input vector, and the authors assume , .
- In their implementation, the centralized observer uses two methods for estimating the team pose: triangulation-based and pair-wise localization.
- The translation between the frames can readily be determined up to a scale factor by applying the sine Fig. 14.
- Fig. 17. Triangular to inline formation switch to avoid obstacles.
A. Hardware Platform
- The cooperative control framework was implemented on the GRASP Lab’s Clodbuster (CB) robots (see Fig. 11).
- Video signals from the omnidirectional camera camera are sent to a remote computer via a wireless 2.4–GHz video transmitter.
- Velocity and heading control signals are sent from the host computer to the vehicles as necessary.
- This reduces the cost and size of the platform.
B. Formation Control
- Initial experiments in formation control were used to validate the dynamic state estimator and corresponding control approach.
- The authors next compared the state observer estimates with the ground-truth position data.
- As an example, in the trial on the left side of Fig. 12, the desired formation was an isosceles triangle where both followers maintained a distance of 1.0 m from the leader.
- Also worth noting is that the actual separation distance of the robots is always greater than desired.
- The controller response is significantly improved as a result of the feedforward terms from the estimator.
C. Switching Formations
- The authors use a simple reactive obstacle avoider [24] on the leader, while allowing the team to choose between either an isosceles triangle or inline convoy formation.
- In the presence of obstacles, the followers switch to an inline position behind the leader in order to negotiate the obstacles while following the leader.
- The internal mode switching in their centralized state observer is also shown in Fig. 18.
- Approximately 3 s into the run, the leader detects the obstacles and triggers a formation switch (triangle to inline).
- The observer mode switches internally from triangular to pair-wise depending on the geometry of the formation.
D. Coordinated Manipulation
- The ability to maintain a prescribed formation allows the robots to “trap” objects in their midst and to flow the formation, guaranteeing that the object is transported to the desired position.
- He received the M.S. degree in mechanical engineering from the Johns Hopkins University, Baltimore, MD, in 1993.
Did you find this useful? Give us your feedback
Citations
697 citations
600 citations
554 citations
503 citations
469 citations
References
15,545 citations
7,291 citations
6,592 citations
3,566 citations
3,008 citations
Related Papers (5)
Frequently Asked Questions (15)
Q2. What have the authors stated for future works in "A vision-based formation control framework" ?
Analyzing the effect of communication constraints, deciding the optimality of formation choices for a given environment, sensor planning for cooperative active vision, and implementing multirobot coordination tasks with a larger number of robots are also important directions for their future work.
Q3. What are the main directions for their future work?
Analyzing the effect of communication constraints, deciding the optimality of formation choices for a given environment, sensor planning for cooperative active vision, and implementing multirobot coordination tasks with a larger number of robots are also important directions for their future work.
Q4. What methods are used for estimating the team pose?
In their implementation, the centralized observer uses two methods for estimating the team pose: triangulation-based and pair-wise localization.
Q5. How can the authors obtain position vectors to within a scale factor?
9. Position vectors relative to other frames can also be obtained to within a scale factor by using the corresponding unit vectors.
Q6. How many robots can solve the pose problem?
the authors note though that when the pose problem is reduced to 2-D space, relative localization can be accomplished by a pair of robots.
Q7. what is the simplest way to show that a robot can reach a certain mode?
The authors need to show that for a given switching strategy , the switched system is stable, i.e., given any initial mode , a desired mode is achieved in finite time.
Q8. What is the need for a switching paradigm for robots?
The authors need a switching paradigm that allows robots to select the most appropriate controllers (formation) depending on the environment.
Q9. What is the role of the controllers in the formation of a robot?
At the coordination level, for an robot formation to maintain a desired shape, the authors need to model the choice of controllers between the individual robots as they move in a given environment.
Q10. What is the way to determine the leader’s trajectory?
any trajectory generated by such a planner for the leader will ensure stable leader-follower dynamics using the above controller.
Q11. What is the kinematic equation for the closed-loop system?
the kinematic equations are given by(10)where is the system output, is the input vector, andOnce again the authors use input–output linearization to derive a control law for which gives us the following closed-loop dynamics:(11)where is an auxiliary control input and is the chosen positive definite controller gain matrix.
Q12. What is the effect of a linearized control graph on the performance of nonholono?
the performance associated with a choice of formation for nonholonomic robots with input–output feedback linearizedcontrollers depends on the length of the path for flow of control information (feedforward terms) from the leader to any follower in the assigned formation.
Q13. What is the simplest way to show a convoy-like system?
If each successive leader’s trajectory satisfies the assumptions of Theorem 1, then the convoy-like system can be shown to be stable.
Q14. What is the simplest way to compute the translation matrix?
From these, the authors obtain the following pairs of equations:(22)With all translation vectors known to a scale factor, the problem of solving for each rotation matrix reduces to the form(23)This can be rephrased as the following optimization problem:(24)The rotation matrix which minimizes this expression can be computed in closed form as follows [34](25)where .
Q15. How can the authors determine the translation between the frames?
With this angle information, the translation between the frames can readily be determined up to a scale factor by applying the sinerule to the shaded triangle in Fig.