This paper treats, in a general way, the problem of mobile robot navigation in a totally unknown environment. The different aspects of this problem are dealt with one by one. We begin by introducing a simple method for perceiving and analyzing the robot's local environment based on a limited amount of distance information. Using this analysis as our base, we present a navigation algorithm containing different action modules; some of these actions use Fuzzy Logic. The results presented whether experimental or simulation show that our method is well adapted to this type of problem.

This paper presents a neural network based control strategy for the trajectory control of robot manipulators. The neural network learns the inverse dynamics of a robot manipulator without any a priori knowledge of the manipulator inertial parameters nor any a priori knowledge of the equation of dynamics. A two step feedback-error-learning process is proposed. Strategies for selection of the training trajectories and difficulties with on-line training are discussed.

A computational technique for designing a physically realizable robust dynamic gait for a planar biped robot is developed. Firstly, a feasible set of gaits was constructed to satisfy the periodicity of the biped locomotion. Then the concept of dynamic, stability margin is introduced based on the robustness of a gait with respect to the external disturbances. Using that margin, we can assess the robustness of each dynamic gait in the feasible set. It is found that the parameter, called foot strike time margin, representing the readiness of the foot strike has a close positive correlation with the dynamic stability margin. We obtain a robust gait with respect to the external disturbance by maximizing the foot strike time margin. The robustness of the optimal gait is confirmed by the behavior of the gait after application of linear impulse as well as by the examination of the largest eigenvalue at the perturbed state.

This paper presents a parametric study of automated vehicles using the sensitivity theory. A sixth order dynamic model of an axisymmetric vehicle is developed in the state space format to represent its 3 degrees-of-freedom motion in the lateral, yaw and roll modes. Variations of the important parameters of the vehicle are grouped into three separate vectors: with the elements consisting of inertia, stiffness and damping, and geometric-kinematic parameters respectively. The effect of every element of these vectors on the state variables is studied carefully, a comparison being made among the state variables to reveal the relative influence of the parameter-induced variations. This helps better understanding which mode of the system is more affected by changes in particular parameters and the severity in the transient response and in the steady-state response. Then the effects of a particular vector on the performance of the system is studied.

A new neural network (NN) control technique for robot manipulators is introduced in this paper. The fundamental robot control technique is the model-based computed-torque control which is subjected to performance degradation due. to model uncertainty. NN controllers have been traditionally used to generate a compensating joint torque to account for the effects of the uncertainties. The proposed NN control approach is conceptually different in that it is aimed at prefiltering the desired joint trajectories before they are used to command the computed-torque-controlled robot system (the plant) to counteract performance degradation due to plant uncertainties. In this framework, the NN controller serves as the inverse model of the plant, which can be trained on-line using joint tracking error. Several variations of this basic technique is introduced; Back-propagation training algorithms for the NN controller have been developed. Simulation results have demonstrated the excellent tracking performance of the proposed control technique.

This paper presents two adaptive schemes for controlling the end-effector compliance of robot manipulators. Each controller possesses a decentralized structure, in which the control input for each configuration degree-offreedom (DOF) is computed based on information concerning only that DOF. The first scheme is developed using an adaptive impedance control approach and consists of two subsystems: a simple “filter” which modifies the end-effector position trajectory based on the sensed contact force and the desired dynamic relationship between the position and force, and an adaptive controller that produces the joint torques required to track this modified trajectory. The second compliant motion control strategy is an adaptive admittance controller for position-controlled manipulators. In this scheme a desired contact force is specified and then position setpoints for the “inner-loop” position controller are generated which ensure that this desired force is achieved. The proposed controllers are extremely simple computationally, do not require knowledge of the manipulator dynamic model or parameter values of the manipulator or the environment, and are implemented in decentralized form.

This paper aims at contributing to a sub-symbolic, feedback-based “theory of robotic grasping” where no full geometrical knowledge of the shape is assumed. We describe experimental results on grasping 2D generic shapes without traditional geometrical processing. Grasping algorithms are used in conjunction with a vision system and a robot manipulator with a three-fingered gripper is used to grasp several different shapes. The altorithms are run on the shape as it appears on the computer screen (i.e. directly from a vision system). Simulated gripper ringer with virtual sensors are configured and positioned on the screen whose inputs are controlled by moving their position relative to the image until an equilibrium is reached among the control systems involved.

The kinematic structure of closed chain mechanisms is suitable for putting an excess number of inputs in addition to the required number to control their position and orientation. With the addition of these excess inputs, it is possible to generate and control the internal forces within the mechanisms by simultaneously activating joints whose action opposes each other. The internal forces balance inside the structure of mechanisms and do not create any resultant force. However, at the static or quasi-static cases they change the stiffness of the mechanisms. In dynamic cases, in addition to the change in stiffness, they may also generate a damping effect. As a result of these effects, the dynamic behavior of closed chain mechanisms changes. This study analyzes the dynamics of the closed chain mechanisms under the effect of internal forces through the simulation and experimental results using a planar closed chain mechanism.

A major problem in inverse kinematics algorithms is that the generated joint velocities to be fed to the joint servos may cause violation of the speed limits of the joint actuators. In this paper, it is shown how to properly cope with joint velocity limits in first-order inverse kinematics algorithms; the proposed technique guarantees tracking of the desired end-effector path. This goal is achieved by suitably slowing down the task-space trajectory when joint velocity limits are encountered. The time law is modified through a time warp such that the introduced virtual time allows fulfillment of the velocity constraints. A case study is developed to show the effectiveness of the proposed method and a kinematic control scheme based on the presented technique is implemented to demonstrate feasibility under real-time constraints.

This paper details the design principles of operation of a pneumatic proximity-to-tactile sensing device for part handling and recognition in a flexible manufacturing environment. The sensing device utilises a densely packed line array of piezoresistive pressure sensors, providing continuous variable outputs. The sensing plane of the device incorporates a corresponding line array of air jets which develop an air cushion when striking a target of interest. The back pressure levels from these air jets form the basis for the task of target detection and recognition.

This paper discusses three methods of training multi-layer perceptrons (MLPs) to model a six-degrees-of- freedom inertial sensor. Such a sensor is designed for use with a robot to determine the location of objects it has to pick up. The sensor operates by measuring parameters related to the inertia of an object and computing its location from those parameters. MLP models are employed for part of the computation. They are trained to output the orientation of the object in response to an input pattern that includes the period of natural vibration of the sensor on which the object rests. After reviewing the working principle of the sensor, the paper describes the three MLP training methods (backpropagation, optimisation using the Levenberg-Marquardt algorithm, evolution based on the genetic algorithm) and presents the experimental results obtained.