This paper studies a path planning method for multiple robots in unknown environment. Multiple robots adopt the leaderfollowing formation method. For the Q-learning algorithm used by the leader robot, the Q-table is initialized by prior information of environment and the idea of filling concave obstacles is proposed. Then the strategy of choosing actions is improved by simulated annealing algorithm, which changes the greedy factor in real time according to the Q-learning. The follower robot uses an improved gravitational potential field method to follow the leader robot. The simulation results show that the improved algorithm is effective and multiple robots can plan an optimum path to reach the destination with this method.

Aiming at the formation and path planning of multirobot systems in an unknown environment, a path planning method for multirobot formation based on improved Q -learning is proposed. Based on the leader-following approach, the leader robot uses an improved Q -learning algorithm to plan the path and the follower robot achieves a tracking strategy of gravitational potential field (GPF) by designing a cost function to select actions. Specifically, to improve the Q-learning, Q -value is initialized by environmental guidance of the target’s GPF. Then, the virtual obstacle-filling avoidance strategy is presented to fill non-obstacles which is judged to tend to concave obstacles with virtual obstacles. Besides, the simulated annealing (SA) algorithm whose controlling temperature is adjusted in real time according to the learning situation of the Q -learning is applied to improve the action selection strategy. The experimental results show that the improved Q -learning algorithm reduces the convergence time by 89.9% and the number of convergence rounds by 63.4% compared with the traditional algorithm. With the help of the method, multiple robots have a clear division of labor and quickly plan a globally optimized formation path in a completely unknown environment.

Many robots utilize commercial force/torque sensors to identify inertial properties of unknown objects. However, such sensors can be difficult to apply to small-sized robots due to their weight, size, and cost. In this letter, we propose a learning-based approach for estimating the mass and center of mass (COM) of unknown objects without using force/torque sensors at the end effector or on the joints. In our method, a robot arm carries an unknown object as it moves through multiple discrete configurations. Measurements are collected when the robot reaches each discrete configuration and stops. A neural network then estimates joint torques from encoder discrepancies. Given multiple samples, we derive the closed-form relation between joint torques and the object's inertial properties. Based on the derivation, the mass and COM of the object are identified by weighted least squares. In order to improve the accuracy of inferred inertial properties, an attention model is designed to generate the weights used in least squares, which indicate the relative importance for each joint. Our framework requires only encoder measurements without using any force/torque sensors, but still maintains accurate estimation capability. The proposed approach has been demonstrated on a 4-degrees-of-freedom robot arm.

No abstract