A hierarchical framework for physical human-robot interaction

Abstract

Robots are becoming more than machines performing repetitive tasks behind safety fences, and are expected to perform multiple complex tasks and work together with a human. For that purpose, modern robots are commonly built with two main characteristics: a large number of joints to increase their versatility and the capability to feel the environment through torque/force sensors. Controlling such complex robots requires the development of sophisticated frameworks capable of handling multiple tasks. Various frameworks have been proposed in the last years to deal with the redundancy caused by a large number of joints. Those hierarchical frameworks prioritize the achievement of the main task with the whole robot capability, while secondary tasks are performed as well as the remaining mobility allows it. This methodology presents considerable drawbacks in applications requiring that the robot respects constraints imposed by, e.g., safety restrictions or physical limitations. One particular case is unilateral constraints imposed by, e.g., joint or workspace limits. To implement them, task hierarchical frameworks rely on the activation of repulsive potential fields when approaching the limit. The performance of the potential field depends on the configuration and speed of the robot. Additionally, speed limitation is commonly required in collaborative scenarios, but it has been insufficiently treated for torque-controlled robots. With the aim of controlling redundant robots in collaborative scenarios, this thesis proposes a framework that handles multiple tasks under multiple constraints. The robot’s reaction to physical interaction must be intuitive and safe for humans: The robot must not impose high forces on the human or react unexpectedly to external forces. The proposed framework uses novel methods to avoid exceeding position, velocity and acceleration limits in joint and Cartesian spaces. A comparative study is carried out between different redundancy resolution solvers to contrast the diverse approaches used to solve the redundancy problem. Widely used projector-based and quadratic programming-based hierarchical solvers were experimentally analyzed when reacting to external forces. Experiments were performed using an industrial redundant collaborative robot. Results demonstrate that the proposed method to handle unilateral constraints produces a safe and expected reaction in the presence of external forces exerted by humans. The robot does not exceed the given limits, while the tasks performed are prioritized hierarchically.