A hybrid technique for path planning of humanoid robot NAO in static and dynamic terrains

Highlights

• Navigation in static and dynamic terrain.

• Implementation of hybrid DWA–TLBO​ technique for obstacle avoidance in humanoid NAO.

• Implementation of Petri-Net controller for avoiding inter-collision.

• Simulation and experiment on NAO humanoid robot in WEBOT and laboratory conditions respectively.

• Comparison between hybrid DWA–TLBO technique with previously implemented technique.

Abstract

The humanoid robot is widely used because of its ability to imitate human actions. The selection of navigational techniques is of prime importance because the quality of the opted technique directly affects the success of output. In this paper, the hybridization of the Dynamic Window Approach (DWA) and the Teaching–Learning-Based Optimization (TLBO) technique and its implementation on the NAO humanoid robot for navigation have been presented. The input is based on the location of obstacles and the target. The parameters are provided to the DWA technique, which decides the optimum velocity. The intermediate result is feed to the TLBO technique, which operates based on the teacher phase and the learner phase. This hybridization provides an optimum angle to take a turn and avoids the obstacles while moving towards the target. The current article concentrates on implementing hybridized techniques in static and dynamic terrains. Single NAO and some random obstacles are chosen for static navigation. For dynamic terrains, multiple NAOs and some static obstacles are considered. In this case, one humanoid robot acts as a dynamic obstacle to another. In the dynamic terrain, there is a possibility of inter-collision amongst NAOs. To avoid inter-collision, a Petri-Net controller has been designed and implemented in all NAOs. Simulation and experimental results on humanoid NAOs demonstrate target attainment with collision-free optimal paths. Experimental and simulated results of the proposed technique present an acceptable relation under 5 % and 6 % for a single robot and multiple robots, respectively. The proposed technique has been compared with previously developed techniques in complex, danger and dynamic terrains. In comparison with previously developed techniques, it is evident that the proposed technique is robust and efficient for the path planning of humanoid robots.

Introduction

Humanoid robots bring forth the challenges in gait and posture analysis due to complications in the modeling of joint torques for their smooth movement. They have been popular amongst various researchers in recent years. A humanoid robot is being extensively used today because it can smoothly imitate human behaviors. It is beneficial in obtaining information about the surroundings and performing the task a human can perform. There are many workplaces and situations where a human cannot reach easily and execute the tasks. The demands of the humanoid robots are expanding because it adapts to any situation in the workplace of humans. Robots are preferred at various industries and research laboratories because it increases the output and simultaneously the efficiency of the work. It also decreases the risks at the workspace. However, manipulators and mobile robots cannot be used in every place for all tasks; therefore, humanoid robots are utilized to execute various tasks. In the present scenario, humanoid robots are replacing humans from different workplaces and increasing productivity. It is used to assemble cars, military tactics, underwater tasks, electronic factories, etc. Hence, humanoid robots are acceptable in various workplaces for different jobs. Navigational techniques are necessary to guide the humanoid robot and perform the assigned tasks. An optimal technique is required to perform the tasks efficiently and avoid the obstacles if detected in between the path in minimum possible time. Researchers have focussed on various AI (Artificial Intelligence) techniques based on their requirements. Path planning has earned the notion to be among the most complex areas of robotic research in recent years. Combined with the humanoid NAO as the robotic platform, the complexity increases many folds, given the posture analysis and joint movement problems in a multi DOF structure, which is much less while using wheeled mobile robots. Various navigation strategies employed in recent years are categorized into classical and artificial intelligence approaches. Based on the problem-solving strategy, the classical approach derives its methodology from statistical techniques, whereas the AI approach is mostly nature-inspired techniques. Few techniques are discussed over here.

Jun et al. [1] have extensively worked on path planning by the implementation of a novel technique that includes two RRT-like techniques. Authors have estimated the robots pose and resolved the Lemke’s method to solve the linear complementarity problem to compute the robot’s stability. Duchon et al. [2] have implemented improved and modified A* technique for the navigation of a mobile robot established on grid maps. They have concentrated on modification of the path and time based on the complexity of terrains. To avoid collisions and to guide the robot towards the target, Ali et al. [3] have proposed a novel technique that utilizes both orientation and current position of other robots. Three different terrains have been utilized to simulate and validate the technique. The idea of using the Bacterial Potential Field has been implemented in [4] to avoid collisions with dynamic and static obstacles. The authors, to examine the technique, have outlined MR realistic model applied simulation platform. The concept of a Fuzzy Logic system has been implemented in [5] for the path planning of mobile robots in static terrains. The authors have simulated and experimented in different terrains to check the efficiency of the proposed concept and computed the angular velocity of the wheels. Patle et al. [6] have achieved smooth navigation in an uncertain terrain by employing the Firefly Approach (FA). They have replaced the static terrain into dynamic and implemented the proposed approach. The proposed technique provides a better result when compared with other path planning approaches. Authors in [7], [8] have presented various AI techniques for path planning mobile robots. Further, the FA has been compared with the Ant Colony Optimization (ACO) technique to ensure the optimality of the path planning technique [9]. The comparison result shows that the FA provides better path length with the lesser computational cost. Bajrami et al. [10] have introduced an Artificial Neural Fuzzy Logic technique to provide the avoidance of collisions with the obstacles and ensure the least processing time. Authors have both simulated and experimented with the proposed technique on a real mobile robot. Various AI techniques have been implemented in a mobile robot for navigation in various complex terrains [11], [12]. Kashyap et al. [13] have described the dynamics of the humanoid robot and further using whole body control and further implemented 3D LIPM [14] for its gait planning. Samant et al. [15] have implemented fuzzy logic in a humanoid robot for proper interaction with the competitive environment. They have validated the results during experimentation during soccer gameplay. To get the optimized design,

Sun and Er [16] have hybridized the Fuzzy Logic with a Genetic Algorithm. Lacevic and Osamkovic [17] have proposed a bur of free C-space to improve the path planning for rigid bodies. Rath et al. [18] have ensured the collision-free path and arrived at the desired destination by the implementation of the Fuzzy Intelligent scheme into action. The authors have compared the simulated and experimented results to ensure the efficiency of the proposed scheme on the humanoid NAO robot. Authors in [19] have explained the high-energy utilization in the operation of the humanoid robots by the implementation of two different methods (Iterative and Fuzzy Logic) for proper gait control of the humanoid robot. Asif et al. [20] have worked on humanoid motion planning using a multi-heuristic search. The search space constraints and complexity are the major disputes in obtaining footstep planning. The authors have worked in various terrains to justify the efficiency of the proposed scheme. Kumar et al. [21] have implemented the hybrid technique in multiple humanoid robots for various scenarios using the Petri-Net controller. Harandi et al. [22] have used deep learning for dimension minimization of the input parameters of the controller alongside maintenance of the performance level and proper execution of the tasks. The visual memory-based technique has been used in humanoid robot NAO for navigation [23]. Muni et al. [24] have implemented Bacterial foraging optimization in the humanoid robot to make it efficient to navigate in complex terrains. The complete analysis of the humanoid robot has been presented in [25]. A non-uniform sampling technique for the humanoid robot has been proposed to get collision-free path planning [26]. It provides safe navigation by avoidance of obstacles in its path. Orgen and Leonard [27] have presented the technique’s convergence property and described the inadequacy in its theoretical treatment. They have proposed convergent DWA to avoid obstacles. An extended version of DWA, which is inspired by Model Predictive Control, has been implemented to control Lyapunov friction and hence, makes the methodology convergent and trackable [28]. Henkel et al. [29] have emphasized the energy-efficient navigation of mobile robots using a Linear Regression model. Molinos et al. [30] have described the adaptation of dynamic window for dynamic tree and dynamic obstacle. They have simulated and experimented in real terrain to approve the efficacy of the proposed technique. Hang et al. [31] have proposed a hybrid technique, where fuzzy logic is used to set the acceptable weight in real-time with DWA. It enhances the efficacy of the original DWA technique, which makes the navigation smoother.

Based on a thorough survey of accessible research papers, it has been observed that path planning in the field of mobile robots is very popular, but these instances are rare in the case of humanoid robots. Humanoid robot navigation is not mainstream. There are a lesser number of existing research papers where path planning has been performed for humanoid robots. And it is also limited for various complexity of terrains, and the application of multiple robots in the single terrain is also an untouched topic as per the author’s knowledge. Various advances have been made to-date in the field of humanoid path planning analysis using AI techniques, but these approaches are mostly concentrated around footstep plans, stability and posture control. Obstacle avoidance using an online computational strategy is uncommon, and the application of a hybrid technique using one of these strategies is yet to be modeled. The proposed research has the prime motive in the modeling, development, and implementation of a hybrid controller for a single cum multi-humanoid system in a single cluttered environment. Due to the limitation of complexity in the path planning of a single robot and multiple robots in single terrain, this paper emphasized on the path planning of a single humanoid robot and multiple humanoid robots in complex terrains. The novel hybrid DWA–TLBO technique has been proposed to obtain the optimized path during path planning in complex terrain. The parameters of TLBO does not require to be tuned; this property enhances the effectiveness of the DWA technique after hybridization. The proposed hybrid technique is capable of planning the path in terrains having a single robot. In the scenario of the presence of multiple robots and to set their prioritization, the proposed technique is not so efficient. Therefore, a Petri-Net model has been implemented to solve the problem of conflicting situations. Various researches on the TLBO technique and Petri-Net control have been presented here.

Khatir et al. [32] have discussed the damage assessment in beam-like structures. They have experimented using steel been using the finite element method (FEM) and isogeometric analysis (IGA). They have performed it in two steps using normalized modal strain energy indicator in the first stage and TLBO in the second stage. Qu et al. [33] have proposed a hybrid BFGS (Broyden–Flecher–Goldfarb Shanno)-TLBO technique to obtain the local and global exploration. Teaching–learning based optimization based on course by course improvement has been proposed [34] to enhance the problem-solving strength of the TLBO technique. An adaptive exponential distribution inertial weight [35] is hybridized with the TLBO technique to solve the problem in a more efficient manner. To solve the premature convergence of the TLBO technique, Xu et al. [36] have preferred a Dynamic-Opposite Learning Approach to hybrid with TLBO. Zhong et al. [37] have described the Decomposition Method for stand-off manage of a class of Petri-Net modeling Flexible Manufacturing Systems (FMS). It efficiently minimizes the computational expenses of plotting a liveness-enforcing Petri-Net manager. Febbraro and Sacone [38] have hybridized the classical Petri-Net model with the fluid version of timed Petri-Net to change the behavior of the system, so that it can adjust with the new situations and still maintains the objectives of control. Holliday and Mary [39] have proposed the generalized timed Petri Net model and describes its performance.

The path planning for single and multiple humanoid robots is inefficient by the application of single AI techniques. Therefore, hybrid AI technique has been proposed because, at an indefinite time, the path planning may also get tricked in local optima and not able to solve the complex path planning problems. To overcome this limitation, the hybridization of the AI techniques has been focussed on this current work. Various path planning researches [40], [41], [42], [43], [44] based on hybridization in the field of mobile robots has been performed, and very fewer research papers are present for path planning of humanoid robots. Following this research breach, a novel hybrid DWA–TLBO technique has been applied in humanoid robots for single and multiple robots. The proposed technique has been designed and implemented in the humanoid robot NAO to obtain optimum travel length and travel time. The breakthroughs achieved via the proposed analysis is described in the given text. In spite of significant achievements and progress in the field of robotics, humanoid robots are still far away from the general perceptions about them as depicted in fictional characters. The demand for the current scenario is a smart navigational system that is able to identify the unknown environment and make decisions based on it. Further addition of a multi-robot system makes the decision making process complex due to the creation of pseudo-dynamic obstacles in the environment. The proposed work emphasizes the modeling and execution of a hybrid DWA–TLBO based controller, which can function efficiently in these situations and can optimize the given conditions to achieve robustness and accurate solutions. Previously developed path planning techniques have some limitations, e.g., long term planning, long term procedural knowledge, and best decision making (target seeking). These limitations have been taken out by the proposed technique, which makes the technique more feasible towards humanoid navigation in comparison to other existing methods. Various experiments and simulations show a coherent behavior among them and correlate to each other on the results achieved. Previous tasks show the indifference shown to the areas of obstacle avoidance and target achievement while designing for movement control of the humanoid. In the presented work, ample progress has been made for the smooth movement of the humanoid robot by providing requires attention to the area of obstacle avoidance in static and dynamic environments.

The DWA technique is based on the velocity, and it is capable of providing optimum velocity. Whereas, TLBO technique has the property of providing an optimum turning angle and does not need tuning of its parameters. TLBO also gets optimized twice based on the teacher phase and the learner phase. The hybridization of these two AI techniques is capable of providing optimized travel time and travel length. The paper is standardized as follows: framework of DWA is described in Section 2, Section 3 describes the framework of TLBO, and Section 4 has introduced the framework of the Petri-Net controller. In Section 5, the proposed hybrid DWA–TLBO technique is presented, and Section 6 discusses the simulation and experimental results. Comparison is performed in Section 7, and finally, Section 8 explains the conclusions and future work.