A bio-inspired kinematic controller for obstacle avoidance during reaching tasks with real robots

Abstract

This paper describes a redundant robot arm that is capable of learning to reach for targets in space in a self-organized fashion while avoiding obstacles. Self-generated movement commands that activate correlated visual, spatial and motor information are used to learn forward and inverse kinematic control models while moving in obstacle-free space using the Direction-to-Rotation Transform (DIRECT). Unlike prior DIRECT models, the learning process in this work was realized using an online Fuzzy ARTMAP learning algorithm. The DIRECT-based kinematic controller is fault tolerant and can handle a wide range of perturbations such as joint locking and the use of tools despite not having experienced them during learning. The DIRECT model was extended based on a novel reactive obstacle avoidance direction (DIRECT-ROAD) model to enable redundant robots to avoid obstacles in environments with simple obstacle configurations. However, certain configurations of obstacles in the environment prevented the robot from reaching the target with purely reactive obstacle avoidance. To address this complexity, a self-organized process of mental rehearsals of movements was modeled, inspired by human and animal experiments on reaching, to generate plans for movement execution using DIRECT-ROAD in complex environments. These mental rehearsals or plans are self-generated by using the Fuzzy ARTMAP algorithm to retrieve multiple solutions for reaching each target while accounting for all the obstacles in its environment. The key aspects of the proposed novel controller were illustrated first using simple examples. Experiments were then performed on real robot platforms to demonstrate successful obstacle avoidance during reaching tasks in real-world environments.

Introduction

The future of industrial robots is progressing towards systems that have a great deal of flexibility in handling various tasks in a self-organized and safe manner. One such task is in the area of intelligent robotic assembly. This task requires the robot to reach and grasp complex objects and manipulate them while satisfying many requirements: the robot must be able to reach for any point in its workspace from any other point with varying speed requirements; the robot must be able to avoid collisions with objects in its environment; it must be able to tolerate unexpected contingencies and perform reliably despite changes in sensor geometry and changes in joint mobility; it must be able to shape its hand around the object to enable stable grasps. In this study the problem of collision avoidance in highly complex environments is developed by drawing inspiration from biological systems.

Previous work has demonstrated obstacle avoidance in complex environments using a path planning approach (Schwarzer & Saha, 2005). In this framework, paths are generated by a path planner, and subsequently evaluated by a separate process to determine collisions with obstacles in the environment (Barraquand et al., 1997, Sanchez and Latombe, 2003). By exhaustively searching through the space of possible paths, a candidate non-collision path is selected for execution. In contrast, path selection and avoidance is an integrated process in human and animal reaching behaviors, and forms the basis of bio-inspired approaches to obstacle avoidance and movement planning.

Another approach to obstacle avoidance is based on modifying path plans created with attractor dynamics (Iossifidis & Schoner, 2007). The controller is developed on the basis of a readily available attractor dynamics model that provides a defined trajectory for the end-effector to reach a target dependent on robot geometry and absolute joint angles. This trajectory serves as the attractor. The influence of the obstacles is superimposed on the attractor dynamics to modify the trajectory. There are three key differences between this approach and the work described in this paper. First, the controller in Iossifidis and Schoner (2007) requires a specific trajectory to be defined in order to reach a target. The trajectory for the novel controller described in this paper is automatically generated by only using directions to targets and thus is less dependent on system parameters and also much easier to implement. Second, the controller in Iossifidis and Schoner (2007) is dependent on absolute joint angles to define the attractor dynamics while the controller described in this work learns a mapping between the changes in the joint angles and changes in direction to the target. Third, the influence of the obstacles modifies the changes in joint angles in this work, while in Iossifidis and Schoner (2007) obstacles modify the absolute joint angles. Thus, the bio-inspired controller described in this paper scales well to robots with more degrees of freedom while being robust to unexpected faults and contingencies.

A prior work that explicitly addresses the obstacle avoidance problem in complex environments is described in Mel (1990). This bio-inspired approach performs a random exploration of arm movements “mentally” using robot models. Each arm configuration that is generated is evaluated for its ability to reach the target while avoiding obstacles. If none of the random movements from the present configuration is successful, then the system “backtracks” to a previous configuration and the search process is continued. This approach creates jerky movements that are arguably unnatural. Furthermore, it is unclear if any subsequent movement smoothing could be applied to avoid collisions.

The approach described in this paper has similarities to Mel (1990) in that “mental rehearsals” (also called lookahead planning) are performed to decide on a plan to avoid obstacles. The mental rehearsal approach employs a perceptual process that is used to generate via-points or intermediate targets during reaching tasks. The via-points are tested for suitability to reach both the target and the initial arm configuration via a reactive obstacle avoidance controller. This approach obviates the need for “backtracking” (Mel, 1990) and is more efficient and practical. Another advantage is that arm movements are smooth due to seamless target switching from via-point to target.

The rest of the paper is organized as follows. Section 2 first describes the basic DIRECT model, followed by a description of the DIRECT-ROAD model for reactive avoidance. Next, the limitation of using the reactive controller in more complex obstacle and target configurations is motivated with an example of a local minima situation. Computer simulations are described to demonstrate obstacle avoidance in complex environments for a planar redundant robot arm. In Section 3, the overall system architecture is described. This is followed in Section 4 by a description of various experiments that were performed on these platforms to demonstrate examples of obstacle avoidance during reaching tasks in complex real-world environments. Section 5 describes related research while concluding remarks are provided in Section 6.