A hybridized regression-adaptive ant colony optimization approach for navigation of humanoids in a cluttered environment

Highlights

• Determination of navigational parameters for humanoid path planning.

• Design of the RA and AACO navigational controllers.

• Design of hybrid RA-AACO controller using the logic of both RA and AACO.

• Design of a Petri-Net system for inter-collision avoidance.

• Testing of the navigational controller in both simulated and experimental environments.

Abstract

Humanoids are preferred over their wheeled counter parts because of their ability to replace human efforts. Navigation and path planning of humanoids is very much important and challenging area of investigation for robotic researchers to enable the humanoids for accomplishing tedious and repetitive tasks. In this paper, a novel hybridization scheme is attempted for the path planning and navigation of humanoids in a cluttered environment. Here, hybridization has been attempted on NAO humanoid robots using regression technique and adaptive ant colony optimization. In the hybridization scheme, the navigational parameters of the humanoids are fed to the regression controller initially in terms of obstacle distances, and the interim output from the regression controller is again fed to the adaptive ant colony optimization controller to obtain the final output. By using V-REP software, navigation simulations are performed, and the simulation results are also tested against real experimental set-up developed under laboratory conditions. The simulation and experimental results reveal that the humanoids are successful in avoiding the obstacles and reach their destinations safely with path optimization. The results obtained from both the environments are compared against each other and are in good agreement with minimal percentage of errors. The navigational controller is tested for both single as well as multiple humanoids, and it works well for both the cases. Finally, the proposed controller is validated against other navigational approaches, and a significant enhancement of results is obtained.

Introduction

With the development of science and technology, robots are becoming an integral part of human life. Among the different forms of robots, humanoids are popular because of their ability to mimic the human behaviour and replace human efforts in repetitive and tedious tasks. Humanoids are very much integrated to workplace in several industries such as medical assistance, manufacturing sectors, industrial automation, etc. To accomplish smooth operation of humanoids in a human workplace, navigation and path planning is very much important. By designing a proper navigational intelligent controller for the humanoids enables them in avoiding the obstacles that are present in the path and negotiate with humans in the workplace. Navigation and path planning approaches are primarily categorized as local or sensor based approaches and global or model based approaches. In sensor based approaches, the robot is unaware of the environmental conditions, and it negotiates with the obstacles by immediate detection and planning. In model based approaches, the robot is aware of the environmental conditions with obstacle locations, and thus it plans the path by mapping the environment from the beginning itself. Similarly, based on the type of obstacles used, path planning can be divided as static path planning and dynamic path planning. In static path planning, only static obstacles are used in the environment, and in dynamic path planning, dynamic obstacles (either in forms of moving obstacles or other fellow robots) are present in the environment. Dynamic path planning is way more challenging than static path planning as needs careful design of the control architecture considering conflicting situations that may arise in deciding the priority among multiple robots when they navigate in a common platform and encounter same obstacle. The use of artificial intelligent algorithms in humanoid navigation is a relatively new area of investigation among robotics researchers. Therefore, the current investigation is devoted towards the use of a hybrid navigation architecture for path planning of humanoids in complex environments. Over the last few decades, humanoid navigation and robotics research have been the centre of attraction for many researchers. Some of them can be cited over here.

Parhi et al. [1], [2], [3] have used several artificial intelligent (AI) approaches for navigation of mobile robots. They have tested the effectiveness of their proposed controllers through proper simulation and experimental environments. Deepak et al. [4], [5], [6] modified the controlling parameters of different intelligent algorithms and observed an enhanced efficiency with the adaptive parameters. Hugel and Jouandeaue [7] developed a 3D LIP model without any torque in the support phase for the walking pattern of humanoid robots. Sadedel et al. [8] incorporated genetic algorithm (GA) to hip and foot trajectories for an offline path planning approach for 2D humanoid robots. Karkowski et al. [9] used A* algorithm and adaptive 3D action set to develop a real time path planning approach for humanoids considering step by step height information. Ido et al. [10] used motion capture data as an input to a view based sequence and analyzed the walking pattern of humanoids. Mohanty and Parhi [11], [12], [13], [14], [15], [16], [17] used fuzzy and cuckoo search based hybrid system for navigation of mobile robots in complex environments and validated the proposed controllers in practical environments. Dalibard et al. [18] developed a collision free path for the dynamic walking of humanoids by a randomized algorithm. Clever and Mombaur [19] proposed an optimal inverse control scheme for motion transfer from humans to humanoids. Mirjalili et al. [20] developed an inverted pendulum model to propose an online path planning approach for SURENA-III humanoid robot. Pradhan et al. [21] used fuzzy logic for navigation of both single and multiple mobile robots with path optimization. Shakiba et al. [22] proposed a modified particle swarm approach by adding Ferguson splines to generate a collision free path for soccer playing humanoids. Perrin et al. [23] compared among different footstep planning approaches applicable to humanoid path planning to generate a classical motion planning approach. Ryu et al. [24] developed a way point based path for humanoids. Parhi et al. [25], [26], [27] developed different adaptive optimization techniques for navigation of mobile and underwater robots. Shimizu and Sugihara [28] used the transitional sequence of the double support phase to propose a path planning approach for the humanoids. Fen et al. [29] generated a collision free path for a humanoid manipulator by modifying the basic ant colony optimization (ACO) algorithm. Kanoun et al. [30] used a virtual kinematic tree as an inverse kinematics problem to generate a path for humanoids. Schmid and Woern [31] generated smooth and collision free path for humanoids by using NURBS curve. Pham and Parhi [32] used neural network as a potential navigation strategy for mobile robots by designing the control architecture considering the environmental constraints. Niskiwaki et al. [33] proposed a laser range finder based path planning approach for humanoids in a complex environment. Yoo and Kim [34] developed a gaze control based architecture for navigation of humanoid robots in complex environments.

The extensive survey of the available literatures suggest that navigation and path planning is very much popular in case of mobile robots. A very few citations are available for humanoid navigation. Although some of the researchers have attempted humanoid navigation up to some extent, most of their approaches are not destination directed, and they apply to limited environments also. Along with that, navigation of multiple humanoids on a common platform has not been reported in any of the literatures as per author’s knowledge. Based on the limitations available in the literatures regarding the navigation of humanoids, the current work is focused on the development of a sensor based navigation strategy for single as well as multiple humanoids in a cluttered environment. To optimize the path followed by the humanoids, a novel hybrid technique comprising of regression analysis (RA) and adaptive ant colony optimization (AACO) has been proposed. Here, the basic control parameters of Ant Colony Optimization (ACO) have been changed to enhance the effectiveness of the original algorithm. Several researchers have tried to control the parameters of the ACO technique, and some of them can be highlighted over here. Castillo et al. [35], [36], [37] proposed a navigation strategy for autonomous mobile robots by modifying the controlling factors of fuzzy logic by ant colony optimization and similarly modifying the factors of ant colony optimization by fuzzy logic. They have experienced a performance improvement by the modification of controlling factors. Zhong and Ai [38] modified the basic ant colony algorithm for balancing a multi-objective assembly line. They observed a minimized workload variation among workstations. Mohammed [39] used a modified approach of ant colony optimization for solution of a travelling salesman problem. Brand et al. [40] used ACO for navigation of mobile robots. They tested their approach in both static and dynamic environments. Purian and Sadeghian [41] used ACO for optimizing the controlling factors of a fuzzy logic and used it in navigation of mobile robots. Wang et al. [42] modified the basic ACO approach by focusing on the solution to pheromone overlapping and used it in a network coding resource optimization. Habib et al. [43] used a combined approach of Voronoi diagram and modified ACO for path planning of mobile robots on point to point motion planning scheme. Krentz et al. [44] proposed a navigation strategy for multiple mobile robots by modifying the basic ACO and considering the challenges involved in shifting the arena from simulation to experimental ones. Han et al. [45] developed a navigation strategy for mobile robots by considering the critical obstacles as the initial pheromone trails in a basic ACO algorithm. By modifying the basic scheme of the algorithm, path length could be optimized as the ants search for the optimal path near the critical obstacles rather than searching the entire space. Reshamwala and Vinchurkar [46] reviewed about the use of ant colony technique in different engineering applications. It can be noticed from the survey of earlier researches that ACO and its modified versions have been mostly applied for mobile robot navigation and industrial applications. The use of the same in humanoid navigation is not yet reported.

The application of a humanoid robot in a complex and dynamic environment demands use of hybrid AI techniques as the individual approaches may not always be self-sufficient for the stated purpose. Sometimes, the navigation problem may also experience trapping in a local optima and unable to deal with navigation of multiple humanoids in a common platform. To avoid the limitations of the standalone methods as navigational approcahes, hybridization is attempted. Several researchers [47], [48], [49], [50], [51] have attempted hybridization in the past for navigation of mobile robots. However, the application of the same in humanoid robots is very rare to find. Based on the above research gap available, the objective of the current investigation is set as the design and implementation of a novel navigational controller that can be used to navigate single as well as multiple humanoid robots in a complex environment with optimization of path and time taken to reach the desired destination. Here, regression analysis is chosen as a classical technique; AACO is chosen as an artificial intelligent algorithm, and a scheme of hybridization is attempted between them to navigate the humanoids. Classical approaches are known to produce converged results within a limited time; however, they are mostly dependent upon the training pattern data and the results produced may lack accuracy in comparison to AI approaches. AI approaches may take somewhat more time to converge; however, they produce more accurate results. Therefore, the hybridization between classical and AI techniques is supposed to take the positive aspects from both the techniques and provided refined outputs.