Intelligent Vision System for Swarm Robotics
A Research Proposal
Your Name
Introduction:
Walter Lippmann said “When all think alike, no one thinks very much”. At first implementing this idea doesn’t look very promising as indeed we always look for innovation and some new ideas but if we broaden our vision a bit more then we would know that swarming a population is not a very bad idea every time. Think of an intelligent population performing search and rescue in some hazardous environment or a ware house perfectly maintained autonomously, isn’t it an incredible idea? The implementation of this concept is realized in the Swarm Robotics. Just like the aunts, bees or similar insects the movement and behavior of each member of a robotic swarm is well coordinated and aligned with each other. The term “Swarm Robotics” was first coined in 1989 by Gerardo Beni in 1989 and he described it as
“The group of robots is not just a group. It has some special characteristics, which are found in swarms of insects, that is, decentralised control, lack of synchronization, simple and (quasi) identical members.”[1].
Swarm Intelligence:
Swarm Robotics basically deals in the controlling of large scale homogeneous multi-robot systems in which the swarm may consist of a few to several hundred simple, compact and homogeneous modules working together to perform macro-level tasks. This provides the benefit of simple and less expensive manufacturing alongside the greater tolerance to fault and obviously many are better than one. Perceptibly for all the individuals to work as one, they need to interact with the environment and locally through a complex and global pattern to attain the goal. This property is called the Swarm Intelligence and it got its roots in the small biological species like aunts and bees. [2]
Intelligent Systems for Swarm Robotics:
There are two basic functions of each individual in a swarm; one is to gather the information from the environment, and the other is to react according to the attained information. The first function means that the robot should have the perception to navigate and identify the objects in the environment they evolve and should share the attained knowledge with other members of the swarm in the similar way as ants gather on a sweet candy. But attaining the ability to observe, learn and recognize the objects in variable viewing conditions is a great challenge in both cognitive research and robotics. This process undertakes the employment of artificial intelligence to evolve basing on the visual perception.
Vision Based Systems:
In the last decade the robotics community has witnessed a strong advancement and some novel works in the field of object detection and recognition like the Viola-Jones Framework, Scale Invariant Feature transform and the Speed up Robust Features [3-5]. Recently for the moving object detection in a video sequence, a Knowledge Based Flexible Edge Matching method was also proposed [6]. Some other methods focused on the cognitive approaches by using some memory based cognitive model for object detection. Quite often these methods are successful in providing high recognition rates in the real time but they all are based on the man-made data bases [7]. These data bases are crucial for the successful implementation of the vision and recognition system but the creation of such data base manually is an extensive project and require skilled human workforce. Another limitation of these systems is that they are mostly formed for the individual intelligent robot. Very less work has been performed for the communication between the individuals of the Robot Swarm System (RSS) and for the communications between the command post and the system.
These shortcomings of the existing systems impede the need for the creation of an intelligent vision system which could be able to observe and learn autonomously in any given environment and share the learned knowledge with the other individuals in the population. Only after this the swarm could be able to behave like a single entity to accomplish the required task. An intelligent vision system should enable the decentralization of the swarm along with robustness and adaptivity.
Mother Nature has always proved to be the best source of inspiration to the problems in every domain and Swarm robotics is the field which as a whole is inspired from the behavior and capabilities of the insects [8]. Among the various mechanisms used the acquisition of data through visual capabilities is most desired in the swarm robotics owing to the availability of small miniaturized visual devices. For this purpose the localization systems are widely studied with the general focus on the internal localization where the robot would estimate the position using the fusing internal sensors which may be of proprioceptive or exteroceptive type. This type of estimation can also be performed if the map of the environment is already built in or it is being formed in real time like in the case of SLAM [9]. The drawback of these system types is the requirement of ground truth or the external positioning reference. The most basic external localization system is the Global Positioning System (GPS) but the GPS has a fundamental restraint of unsuitability for indoors owing to the unavailability of the signals. Owing to this limitation of the global system several other designs of localization principles were proposed which can broadly be divided in to Active and Passive types.
There are several technologies reported for the active type for example a 6DoF localization system is proposed comprising on a camera and four LED’s. In this system the markers are tracked after detection in images making the system robust and increasing the performance [10]. One other active approach proposed is the North-Star system which uses the projections of the ceiling as the temporary ambient markers and thus by projecting a known pattern, the position can be obtained by re-projection [11]. The most recent approach which is used for the localization is the ViCon’s Commercial motion capture system which utilizes the high resolution and high speed cameras along with the strong infrared emitters [12]. Despite all the efficiency this system is a very expensive solution so the need of low cost localization system is still there.
To reduce the overall cost and the complexity of the system, several passive vision based methods are recently proposed in the literature. Some of these works make the use of Augmente Reality (AR) which allows the acquirement of the pose alongside the additional information like the ID of the target. In these methods the software libraries are widely used like the ARToolKit, ARToolKit+ and ARTag. Most of the recent works in this strategy report the detection and confusion rates but not the precisions [13-15].
An intelligent vision system allows the swarm to understand their surroundings, act based on that information and also learn from the experience gained through that action. This technology can be vital for both the consumer and industrial applications as they through this technology; they can not only detect and differentiate between different objects but also can interact with them in any way.
The incorporation of intelligent vision system enables the swarm system to learn and adapt to the variable environments and act according to the situation. This is the quality which differentiates it from the traditional robotics as they are programmed to repeat only some specific tasks. The incorporation of intelligent vision system enables them to perform tasks in the case of variable chain of events just like a human. They can observe the actions performed by any subject that may be human or a machine and can exactly replicate them. They can decide by themselves according to the artificial intelligence system about the restrictions of the environment and the final point up-till which they need to perform that task and when to stop. The intelligent vision system incorporating the artificial intelligence enables the swarm to learn quickly and efficiently from the environmental stimulus.
Despite the recent high focus in the field there are many research spaces which still need to be filled like the demonstration learning cannot be easily transferred to the other member accurately. The other methods like the use of sampling and optimization can solve this problem, but they are time consuming and arduous. I plan to visit _____________ to not only earn my degree in this field but to also to use my thirst of knowledge regarding the topic and to put my share in making this world a better place to live in.
References:
[1]. Beni, Gerardo. "From swarm intelligence to swarm robotics." In International Workshop on Swarm Robotics, pp. 1-9. Springer, Berlin, Heidelberg, 2004.
[2]. Miner, Don. "Swarm robotics algorithms: A survey." Report, MAPLE lab, University of Maryland (2007).
[3]. Bay, Herbert, Andreas Ess, Tinne Tuytelaars, and Luc Van Gool. "Speeded-up robust features (SURF)." Computer vision and image understanding 110, no. 3 (2008): 346-359.
[4]. Lowe, David G. "Object recognition from local scale-invariant features." In Computer vision, 1999. The proceedings of the seventh IEEE international conference on, vol. 2, pp. 1150-1157. Ieee, 1999.
[5]. Viola, Paul, and Michael J. Jones. "Robust real-time face detection." International journal of computer vision 57, no. 2 (2004): 137-154.
[6]. Hossain, M. Julius, M. Ali Akber Dewan, and Oksam Chae. "A flexible edge matching technique for object detection in dynamic environment." Applied Intelligence 36, no. 3 (2012): 638-648.
[7]. Wang, Yanjiang, and Yujuan Qi. "Memory-based cognitive modeling for robust object extraction and tracking." Applied intelligence 39, no. 3 (2013): 614-629.
[8]. Camazine, Scott. Self-organization in biological systems. Princeton University Press, 2003.
[9]. Thrun, Sebastian. "Probabilistic robotics." Communications of the ACM 45, no. 3 (2002): 52-57.
[10]. Breitenmoser, Andreas, Laurent Kneip, and Roland Siegwart. "A monocular vision-based system for 6D relative robot localization." In Intelligent Robots and Systems (IROS), 2011 IEEE/RSJ International Conference on, pp. 79-85. IEEE, 2011.
[11]. Yamamoto, Yutaka, Paolo Pirjanian, M. Munich, E. DiBernardo, L. Goncalves, J. Ostrowski, and N. Karlsson. "Optical sensing for robot perception and localization." In Advanced Robotics and its Social Impacts, 2005. IEEE Workshop on, pp. 14-17. IEEE, 2005.
[12]. Mellinger, Daniel, Nathan Michael, and Vijay Kumar. "Trajectory generation and control for precise aggressive maneuvers with quadrotors." The International Journal of Robotics Research 31, no. 5 (2012): 664-674.
[13]. Fiala, Mark. "Artag, an improved marker system based on artoolkit." National Research Council Canada, Publication Number: NRC 47419 (2004): 2004.
[14]. Fiala, Mark. "Comparing ARTag and ARToolkit Plus fiducial marker systems." In Haptic Audio Visual Environments and their Applications, 2005. IEEE International Workshop on, pp. 6-pp. IEEE, 2005.
[15]. Bošnak, Matevž, Drago Matko, and Sašo Blažič. "Quadrocopter hovering using position-estimation information from inertial sensors and a high-delay video system." Journal of Intelligent & Robotic Systems 67, no. 1 (2012): 43-60.