SLAM for Humanoid Multi-Robot Active Cooperation Based on Relative Observation

Abstract

Abstract:The simultaneous localization and mapping (SLAM) of robot in the complex environment is a fundamental research topic for service robots. This paper presents a new humanoid multi-robot SLAM mechanism that allows robots to collaborate and localize each other in their own SLAM process. Each robot has two switchable modes: independent mode and collaborative mode. Each robot can respond to the requests of other robots and participate in chained localization of the target robot under the leadership of the organiser. We aslo discuss how to find the solution of optimal strategy for chained localization. This mechanism can improve the performance of bundle adjustment at the global level, especially when the image features are few or the results of closed loop are not ideal. The simulation results show that this method has a great effect on improving the accuracy of multi-robot localization and the efficiency of 3D mapping.

Active SLAM: A Review on Last Decade

“…In addition to these parameters, AC-SLAM parameters may include (a) the parameters presented by the authors of [86,87], incorporating the multirobot constraints induced by adding the future robot paths while minimizing the optimal control function (which takes into account the future steps and observations) and minimizing the robot state and map uncertainty and adding them into the belief space (assumed to be Gaussian); (b) parameters relating to exploration and relocalization (to gather at a predefined meeting position) phase of robots as described by [88]; (c) 3D mapping info (OctoMap) used by the authors of [89]; and (d) path and map entropy info, as used in [90], and relative entropy, as mentioned in [91]. Relative observation between agents [88] N Localization utility, information gain, cost of navigation [93] N Visual features, map points [94] Weak edges in pose graphs of target agents [95] Frontier points and map information [96] N Localization utility, information gain, cost of navigation [89] N Visual features, optimized paths [90] N Pose and map entropy, Kullback-Leibler divergence [91] Relative pose entropy [97] Visual features, chained localization [87] Multirobot belief evolution by incorporating mutual observations and future measurements [98] N Frontier points and frontier-to-robot distances [99] N Frontiers and relative-position estimates [100] N, Entropy and future measurements [101] N, Information vector and information matrix 1 Centralized. 2 Distributed.…”

“…This communication may be centralized, decentralized, or distributed. In a centralized communication network as presented by the authors of [86,87,91,94,95,97], all the communication between the nodes is routed through the central server. If the central server becomes unavailable/out of service/out of range, then communication is broken, which makes this topology highly vulnerable to communication loss in the case of server failure.…”

“…Since there is dense communication between nodes, managing the communication bandwidth, task allocation, and data packet reduction considerations are the parameters that need to be optimized, as presented in the work of [102]. Typical application scenarios include collaborative localization [87,92,94,97], exploration and exploitation (revisiting already-explored areas for loop closure) [15,96], and collaborative trajectory planning/trajectory optimization [90,91,95]. In the following sections, we discuss these application scenarios in the selected literature.…”

“…The proposed objective function takes into account the evolution of predicted measurements over the planning horizon and the trace of the covariance matrix associated with the robot-pose uncertainty. In an interesting approach, the method presented in [97] uses multiple humanoid robots, where each robot has two working modes, independent and collaborative. Each robot has two threads running simultaneously: (a) the motion thread and (b) the listening thread.…”

Modeling Chinese Secondary School Students’ Behavioral Intentions to Learn Artificial Intelligence with the Theory of Planned Behavior and Self-Determination Theory

“…SLAM refers to a technology for autonomously constructing a map of an unknown environment while positioning the robot on this map [24]. The algorithm is a fundamental mechanism to tackle the positions autonomously and flexibly, and is available for application in different contexts, such as active cooperation and localizing each other, and cooperating in a 3D mapping service [25]. Students could be expected to learn the SLAM algorithm in robotics.…”