Obstacle avoidance for robotic manipulators can be challenging when they operate in unstructured environments. This problem is probed with the sim-to-real (sim2real) deep reinforcement learning, such that a moving policy of the robotic arm is learnt in a simulator and then adapted to the real world. However, the problem of sim2real adaptation is notoriously difficult. To this end, this work proposes (1) a unified representation of obstacles and targets to capture the underlying dynamics of the environment while allowing generalization to unseen goals and (2) a flexible end-to-end model combining the unified representation with the deep reinforcement learning control module that can be trained by interacting with the environment. Such a representation is agnostic to the shape and appearance of the underlying objects, which simplifies and unifies the scene representation in both simulated and real worlds. We implement this idea with a vision-based actor-critic framework by devising a bounding box predictor module. The predictor estimates the 3D bounding boxes of obstacles and targets from the RGB-D input. The features extracted by the predictor are fed into the policy network, and all the modules are jointly trained. This makes the policy learn object-aware scene representation, which leads to a data-efficient learning of the obstacle avoidance policy. Our experiments in simulated environment and the real-world show that the end-to-end model of the unified representation achieves better sim2real adaption and scene generalization than state-of-the-art techniques.

Motivated by the demand of intelligent control that applicable to real industrial systems, this paper develops an innovative neuro controller for remotely operated robotic manipulators. First, the leader-following robotic devices and their communication are introduced. The remotely operated robot can either be controlled by pre-designed controller or humancomputer interaction. Then, a neuro network algorithm is utilized to update PID control gains adaptively. To minimize training time, adaptive moment estimation algorithm is integrated to the neuro PID controller to form a novel control strategy which can be implemented in closed loop. As a result, intelligent algorithms and control methodology can be combined to solve industrial problem. Furthermore, the proposed intelligent controllers are implemented to two cases of remotely operated robot: 1. predesigned motion driven for comparison; 2. Human-computer interaction. Real experimental results can successfully validate the effectiveness of the presented techniques.

A resilient robot can recover its original function after partial damage of the system. This paper develops an underactuated resilient robot that utilizes a combination of passive joints, active joints, adjustable links, and passive links. A novel method based on the genetic algorithm was proposed to determine the goal configuration of a partially damaged robot. The novelty of the method lies in the integration of both discrete and continuous variables. This model is illustrated by a 3-DOF robot manipulator in the simulation.

This paper addresses the problem of relative position-based formation planning for a leader–follower multi-robot setup, where the robots adjust the formation parameters, such as size and three-dimensional orientation, to avoid collisions and progress toward their goal. Specifically, we develop a virtual sub-target-based obstacle avoidance method, which involves a transitional virtual sub-target that guides the robots to avoid obstacles according to obstacle information, target, and boundary. Moreover, we develop a changing formation strategy to determine the necessity to avoid collisions and a priority-based model to determine which robots move, thus dynamically adjusting the relative distance between the followers and the leader. The backstepping-based sliding motion controller guarantees that the trajectory and velocity tracking errors converge to zero. The proposed robot navigation method can be employed in various environments and types of obstacles, allowing for a formation change. Furthermore, it is efficient and scalable under various numbers of robots. The approach is experimentally verified using three real mobile robots and up to five mobile robots in simulated scenarios.