Sensorimotor Control Strategies for Tactile Robotics

TL;DR

This paper surveys tactile robotics with a focus on sensorimotor control for perception and manipulation. It surveys tactile sensing modalities, multi-modal integration, and closed-loop control strategies, including active exploration, tactile servoing, and friction-aware grasping. It highlights challenges such as high-dimensional tactile data, sensor heterogeneity (rigid vs soft), and the need for real-time feature extraction and robust planning under uncertainty. The authors offer practical design guidelines and identify future directions in tactile hardware, learning-based perception, and integrated control architectures to enable more dexterous, compliant, and robust tactile manipulation.

Abstract

How are robots becoming smarter at interacting with their surroundings? Recent advances have reshaped how robots use tactile sensing to perceive and engage with the world. Tactile sensing is a game-changer, allowing robots to embed sensorimotor control strategies to interact with complex environments and skillfully handle heterogeneous objects. Such control frameworks plan contact-driven motions while staying responsive to sudden changes. We review the latest methods for building perception and control systems in tactile robotics while offering practical guidelines for their design and implementation. We also address key challenges to shape the future of intelligent robots.

Figures (8)

• Figure 1: Sensors measure the robot's state and contacts with the environment, through a forward mapping and a contact estimator respectively. Such information is primarily sent to a reactive controller, which can generate a motion correction command for the robot to quickly respond to control stability or robustness losses when contact is achieved. The command is sent to the robot's actuation system through low-level controllers. Similarly, post-processed sensing information is sent to a high-level controller, which allows to plan the manipulation task, making also use of robot models and a memory of past system states. The trajectory is then sent to an actuation command generator specific to the robotic system under examination, which is located within a mid-level block and is added to the previously generated correction.
• Figure 2: The manipulation process involves several distinct stages. Before initiating the grasp, the gripper adjusts its fingers to facilitate future contact sensing and explores the objects in its surroundings. Once a secure grasp is achieved, the gripper firmly holds the object to prevent slippage and enables in-hand manipulation. Finally, the object is placed onto a designated target configuration.
• Figure 3: Tactile sensorization of robotic end-effectors: two-fingered, fin-ray, anthropomorphic, tentacle, and universal. Areas devoted to different sensing have been highlighted with different colors: proximity (red), target properties - texture, stiffness, etc. - (green), pressure and force (blue), and slip (yellow).
• Figure 4: Local and global workspace exploration. Local exploration relies on tactile sensing to gather object surface properties - like its curvature, texture, and roughness. Combined with vision and proprioception, it facilitates higher-level perception like object detection and shape analysis. Global exploration utilizes previously acquired information to make perception-informed decisions, such as object recognition and spatial reasoning.
• Figure 5: (a) The general tactile servo controller. Starting from the current tactile image $S_a$ to reach a goal contact state $C_d$, the tactile servo controller outputs how much the contact should change $\Delta C$. (b) The inverse sensor model maps the features $f_a$ extracted from $S_a$ into the current contact $C_a$; (c) the forward sensor model links the desired contact to the expected tactile image $S_d$, whose feature vector $f_d$ is computed from. It is then compared to the current feature $f_a$, and the contact motion is generated by mean of the Tactile Jacobian and the residual feature vector $\Delta f$.