We present a hybrid learning and model-based approach that adapts internal grasp forces to halt in-hand slip on a multifingered robotic gripper. A multimodal tactile stack combines piezoelectric (PzE) sensing for fast slip cues with piezoresistive (PzR) arrays for contact localization, enabling online construction of the grasp matrix. Upon slip, we update internal forces computed in the null space of the grasp via a quadratic program that preserves the object wrench while enforcing actuation limits. The pipeline yields a theoretical sensing-to-command latency of 35-40 ms, with 5 ms for PzR-based contact and geometry updates and about 4 ms for the quadratic program solve. In controlled trials, slip onset is detected at 20ms. We demonstrate closed-loop stabilization on multifingered grasps under external perturbations. Augmenting efficient analytic force control with learned tactile cues yields both robustness and rapid reactions, as confirmed in our end-to-end evaluation. Measured delays are dominated by the experimental data path rather than actual computation. The analysis outlines a clear route to sub-50 ms closed-loop stabilization.
Grasping and manipulation are fundamental capabilities for robotic hands and grippers, enabling autonomous systems to physically interact with their environment. A central challenge is detecting and controlling in-hand object slip while maintaining grasp stability under external disturbances such as gravity, inertial loads, or unexpected perturbations. This is particularly relevant for fragile or deformable objects, where excessive contact forces can cause damage [1]. Maintaining the lowest feasible grasp forces also improves dexterity, responsiveness, and manipulation precision. Reactive slip control (RSC) strategies that uniformly increase grasp forces upon slip detection can be effective for simple parallel-jaw grippers. However, when applied to multifingered hands, such uniform force increases can introduce undesired objectlevel wrenches, perturbing the object pose and complicating slip recovery. Grasp stability depends critically on coordinated force distribution across multiple contacts. More advanced approaches rely on explicit modeling of frictional interfaces and external forces, but information of friction coefficients, object properties, or disturbance magnitudes is often unavailable or unreliable in real-world settings. To address these limitations, this paper proposes a hybrid datadriven and model-based approach for adaptive grasp force control in multifingered robotic grippers. This research was supported by TraceBot project. TraceBot has received funding from the European Union's H2020-EU.2.1.1. INDUSTRIAL LEADERSHIP programme (grant agreement No 101017089) 1 Université Grenoble Alpes, CEA, Leti, F-38000 Grenoble, France 2 Université Paris-Saclay, CEA, List, F-91120, Palaiseau, France theo.ayral@gmail.com Our approach closes the loop between tactile slip perception and analytic force redistribution: when slip is detected, we adjust internal forces, preserving the object-level wrench, using contact information estimated online from tactile sensing. Integrating piezoelectric (PzE) and piezoresistive (PzR) sensors, our method continuously monitors the contact state and dynamically adjusts internal forces to prevent slippage. The system is designed to optimize grasp equilibrium in real time, ensuring robustness to unknown perturbations. Contributions:
• We develop an adaptive grasp control strategy that abstracts away from manipulation forces (for object motion) and optimizes internal forces (for stability), implicitly enlarging friction margins while preserving the object-level wrench. • We introduce a hybrid tactile pipeline that fuses fast slip cues (PzE) with contact localization (PzR) to update the grasp model online, and we gate execution with simple feasibility checks that determine when internalforce control is applicable. • We validate the approach on multi-finger precision grasps under external perturbations, with a focused analysis of performance and limitations. Section II reviews existing approaches to reactive slip control, grasp-force optimization, and tactile slip sensing, and positions our hybrid strategy. Section III presents the grasp modeling framework together with the reactive slip control pipeline, integrating learned slip detection with internalforce optimization. Section IV details the experimental setup and evaluation protocols. Section V reports experimental validation across multiple grasp configurations, demonstrating stabilization under external perturbations. Section VI discusses applicability and limitations.
Reactive Slip Control (RSC) refers to the process of increasing grasp forces to counteract external disturbances when slippage is detected [2], [3]. This approach to grasp stability relies on tactile feedback in real-time for slip detection. Typically, the increase in force is chosen arbitrarily, with the same force applied by each finger. This method can be used with various types of grippers [1], and the number of fingers is irrelevant as long as the gripper is symmetrical. In such cases, the forces applied by all fingers usually balance out. However, this simple approach is limited to basic grippers. For dexterous grasps involving articulated multi-fingered grippers or robotic hands, a more tailored force distribution across contact points is required. If finger forces are not coordinated, the resulting net wrench can cause uncontrolled motion of the object, compromising the goal of grasp stabilization [4]. For multi-fingered grippers, reactive slip control has also been implemented at the finger level, each finger acting autonomously when detecting object slippage [1], [2]. The work of [5] investigates a probabilistic model for the estimation of grasp stability and increase grasp stiffness if the grasp is estimated to be unstable. If this is not enough, regrasping is performed. In [6], the authors tackle slippage control by explicitly utilizing a friction model. This approach requires direct measurement or estimation of contact forces and torsional momen
This content is AI-processed based on open access ArXiv data.