Reactive Whole-Body Control: Dynamic Mobile Manipulation Using a Large Number of Actuated Degrees of Freedom

Abstract

In this article, we present a control framework for reactive mobile manipulation of robotic systems with a large number of actuated degrees of freedom (DOF). We apply the concept to the humanoid robot Rollin' Justin of the German Aerospace Center (DLR). As service robotics is expected to be established in households and human environments in the near future, we consider relevant aspects like safety, compliance and robust task execution. The multi-DOF manipulator achieves an interactive redundancy resolution while planning algorithms only have to be applied to the low-dimensional operational space concerning task execution. Various experiments have been conducted, e.g., on reaching of a remote object, human-robot interaction, and self-collision avoidance of the manipulator. The results can serve as an interface to (re-)planning methods. Thanks to its interactivity, the approach can be applied in dynamic environments.

Whole-Body Impedance Control of Wheeled Humanoid Robots

“…In the context of impedance control, the minimization of the quadratic norm of a lower-priority impedance error in a two-level hierarchy has been treated in [PAW10]. Multi-objective whole-body compliance control concepts utilizing joint torque sensing have recently been developed [DWASH12b,DWASH12a]. Several works in the literature also address prioritized multi-task control at the kinematic level.…”

“…If there are more DOF available, this remaining kinematic redundancy can be used to pursue further important objectives such as collision avoidance, observation of the environment, or pose optimization for increased energy efficiency. In general, these control tasks have a reactive nature, i. e. the robot is able to locally react on disturbances, unmodeled dynamics, and unpredictable environments to achieve the goals in real time [DWASH12b]. Certainly the most frequently used method to define a reactive control task is to apply artificial attractive or repulsive potential fields [Kel29,Kha86] and to use their gradients as control inputs τ ∈ R n .…”

“…Compared to other weighting matrices, this choice is computationally very cheap. Moreover, the null space can be interpreted from a geometric point of view [DWASH12b], and damped least-squares techniques can be applied easily [DW95].…”

Towards Autonomous Planetary Exploration

“…By using this linearization in an iterative algorithm, the robot can search the configuration space for possible motions while respecting all constraints. -Hierarchical composition of constraints: To combine the individual task constraints in a defined manner, we prioritize the constraints based on their importance, similar to the concept of [25]. We enforce these priorities by projecting low-priority constraints into the null space of the linearization of high-priority constraints.…”

Task-Space Modular Dynamics for Dual-Arms Expressed through a Relative Jacobian

“…For dual-arms (as shown in Fig. 1) and humanoids, the DOFs are: 25 for Compliant Humanoid (CoMan) [1], 34 for Honda Asimo's P2 [2], 41 for iCub upper body [3], 43 for DLR's dual-arm [4], and 51 for mobile humanoid Rollin' Justin [5], to name a few.…”