From multi-modal tactile signals to a compliant control

Abstract

Abstract-In this paper, we present an end-to-end approach transforming multi-modal tactile signals into a compliant control to generate different dynamic robot behaviors. This is obtained by fusing multi-modal sensor signals from our artificial skin and joint sensors with different control approaches. One advantage of these compliant behaviors is to produce safer robots, especially for physical Human-Robot Interaction. A key component of our framework is a robot parametric modeling based on the artificial skin multi-modal sensors (proximity, force and acceleration). These generated models are used to control a robot improving and even changing its dynamic behavior. We validate our framework in a real wheeled humanoid robot, where our presented framework enables a stiff robotic system to be compliant and react to multi-modal tactile events (pre-contacts and contacts).

Tactile-Based Whole-Body Compliance With Force Propagation for Mobile Manipulators

“…Instead it must be guaranteed for an arbitrary number of contact points distributed throughout the entire body of the considered robot. Therefore, such a robot must either have a soft mechanical structure, associated with a proper actuation strategy, e.g., variable stiffness [10], [20], [21], or must otherwise be equipped with specific sensor modalities allowing suitable reactive behaviors to be emulated by its controller [4], [22]. On the other hand, the generation of compliant behaviors must be guaranteed despite the kinematic and dynamic limitations of the considered robot, such as joint limits, torque limits or singularities.…”

“…The artificial skin developed in [45], [46] is of particular interest as it is capable of simultaneously measuring four different tactile modalities, namely pressure, pre-touch, temperature and vibrations. Mounted on a robot, this artificial skin can be used for active compliance realization as demonstrated in [22]. In our previous work [48], we took advantage of the pressure and pre-touch modalities in order to achieve whole-body compliance with force propagation on a dual-arm omnidirectional mobile manipulator.…”

“…(d): Shape reconstruction of a given skin patch can be achieved after a short self-organization of the cells network followed by a self-calibration step [45]. (e): Once the shape of a skin patch is known, each skin cell can be provided with a specific position vector 0 x σ ∈ R m , defined in the base frame of each arm [22]. Notice the corresponding coordinate frames on the left arm of the robot model.…”

“…3) Description of the control interface: TOMM's arms are velocity-controlled (similarly to most industrial robots). However, model-based force/torque control strategies can still be implemented through a set of admittance controllers, converting the generated torque commands into suitable velocity commands as demonstrated in [22]. The control loop frequency is here fixed to 125Hz by the real-time interface used to communicate with each UR5 robot arms 4 .…”

“…We then proceed by gradually covering the robot's artificial skin with a set of opaque screens, in order to activate as many proximity sensors as possible. Since the skin's proximity feedback is automatically mapped to a virtual force [22], it is therefore naturally integrated into our controller. The experiment results are shown in Fig.…”