A Self-organizing Multi-agent Cooperative Robotic System: An Application of Cohort Intelligence Algorithm
This paper presents an application of the emerging Cohort Intelligence (CI) algorithm in the domain of swarm robotics. The application could be relevant to search and rescue in alien territory as well as establishment. The robots are considered as candida
- PDF / 698,138 Bytes
- 14 Pages / 439.37 x 666.142 pts Page_size
- 38 Downloads / 163 Views
28
P. Roychowdhury et al.
1 Introduction According to Grayson [6], Search and Rescue (SAR) robotic systems are becoming important especially in the urban and densely populated environment. The collapsed structures, unknown establishments and hostile environment due to disasters like earthquake, tsunami, etc. add further complexity to the search and rescue operations. A single robotic system for such operation may have certain limitations including less robustness, i.e. such system is prone to single point failure. Moreover, the cost may increase with increase in number of capabilities of the robots. Multi-robotic systems are comparatively more fault tolerant with reduced communication load as well as may bring flexibility in the system [14, 15]. Although vision-based robotic systems have been deployed, most of the systems follow simple coordination rules without explicit teamwork models or goals. An Urban Search And Rescue (USAR) robotic system was developed by Burion [2]. As human intervention is required for the decision making, it adds communication overload and associated delays. This makes the integration and coordination amongst the robots remain a challenge. The algorithm of Cohort Intelligence (CI) was proposed by Kulkarni et al. in [9]. It is inspired from the social behavior of candidates in a cohort. These candidates compete and interact with one another to enrich their independent behavior. Based on certain probability every candidate iteratively chooses another candidate to follow and chooses the values of variables/qualities in the close neighborhood from within. This process continues until there is no significant change in the behavior of all the candidates in the cohort or the goal is achieved. So far CI has been applied for solving continuous unconstrained [9] and constrained test problems [12, 13]. A modified version of CI (MCI) with improved local search ability using mutation was also developed and applied for solving several clustering problems. A hybridized version of CI and K-means was also successfully developed solving these problems [7]. The CI algorithm was further applied for solving large sized combinatorial problems from healthcare domain, a practical version of multiple knapsack problem referred to as sea-cargo mix problem (more than 25,500 variable), and selection of cross-border shipper’s problem from transportation domain (more than 850,000 variables) [8, 10]. In addition, CI was applied successfully for solving 0–1 knapsack problems [11]. It is important to mention that in these earlier versions of CI, every candidate employs linear probability approach to choose a candidate to follow. The linear probability value is directly proportional to the behavior of the candidate and the probability stake associated with the roulette wheel is directly proportional to the quality of the behavior of the individual candidate. There are chances that a worse candidate may be followed by a candidate with comparatively better behavior. According to Kulkarni et al. [9], it helped the candidates
Data Loading...
