Multi-robot Active SLAM


The main motivation of this research is to utilize multi-robot systems to explore indoor and outdoor environments where the accurate global localization method is not always available. In the absence of sufficiently accurate global localization methods, the mapping problem should be solved coupled with the localization problem. This problem is called simultaneous localization and mapping (SLAM).

The original SLAM systems, both in single- and multi-robot systems were passive systems. The influence of the robot on the SLAM algorithm performance was ignored. In the active SLAM systems, the control signal optimizes an objective function related to the SLAM algorithm. The objective function can be chosen to minimize the time, or uncertainty of the SLAM process. The active SLAM has been extensively discussed for the single robot systems, but active SLAM is considered a new topic for the multi-robot system, especially in the vision-based systems.

Two main problems in multi-robot active SLAM is Multiagent Exploration and motion planning in uncertain environments. During the past few years, we have carried out different projects on these two topics in ACIS laboratory.


Multiagent Exploration Strategy Using Consensus-based Auction Algorithm:

In this project, the next-best-view (NBV) exploration strategy has been extended to multiagent robotic systems. The NBV strategy is composed of three steps; first, some random candidate points are chosen from the frontier region, then the reward of moving to each candidate point is calculated for different robots. This reward is a function of the travel cost and the area discovered by moving to that point. An allocation algorithm is used to maximise the sum of the achieved reward over the entire group of agents. Apparently, the reward of each point is a function of the allocation of other points to the other robot. In order to solve the allocation problem, a modified version of the consensus-based auction algorithm (CBAA) has been used. The CBAA is composed of two phases: bidding and consensus based winner determination. By utilizing CBAA, the robot fleet is able to solve the task allocation and find a conflict free consensus in a distributed way with no need of a central auctioneer. An example of the multiagent exploration NBV strategy is shown in the Figure below.

Motion planning in uncertain environment:

Motion control of mobile robots in dynamic uncertain environments is a challenging task. This project is focused on an unscented predictive motion control (UMPC) algorithm to tackle this problem. The main contribution of this paper is utilizing the statistical linearization instead of the analytical linearization to approximate the belief space evolvement. To achieve a Gaussian belief space, the state evolution is approximated using an unscented transform. This approximation enables us to utilize the closed solutions which are available for the linear systems. A model predictive motion control scheme is used to find the suboptimal control policies. In addition to the nonholonomic constraints, state estimation and collision avoidance chance constraints are incorporated to the predictive scheme. Examples of the performance of the proposed UMPC algorithm for single- and multi-robot system are shown in the figures below.


  1. M. Farrokhsiar and H. Najjaran, “An unscented predictive control approach to formation control of multiple nonholonomic mobile robots,” in The 2012 IEEE International Conference on Robotics and Automation (ICRA 2012)
  2. M. Farrokhsiar and H. Najjaran, “Multiagent exploration strategy using consensus-based auction algorithm,” in Proceedings of the Canadian Society for Mechanical Engineering International Congress 2012, 2012
  3. M. Farrokhsiar and H. Najjaran, “Unscented predictive motion planning of a nonholonomic system,” in 2011 IEEE International Conference on Robotics and Automation (ICRA 2011), 2011
  4. M. Farrokhsiar and H. Najjaran, “Formation Control of Nonholonomic Mobile Robots in Unstructured Environments: An Unscented Model Predictive Control Approach,” Journal of Robotica, Submitted
  5. M. Farrokhsiar and Homayoun Najjaran, “Motion Control of Chance Constrained Nonholonomic Systems using an Unscented Model Predictive Approach,” IET Control Theory & Applications Submitted