Tactile Memory for Continuous Policy Blending in Unified Force-Impedance Control
Robin Jeanne Kirschner,, Hamid Sadeghian, and Sami Haddadin
AI summary
Problem
Contact-rich industrial tasks typically rely on rigid state machines with fixed transition conditions, making them tedious to program and poorly adaptable to variations in geometry or position.
Approach
The system uses 32 process quality metrics (PQMs) processed by a BiLSTM to predict operation success, generating soft blending weights for smooth transitions within a unified force-impedance control loop protected by an energy tank for safety.
Key results
- Successful execution of peg insertion, USB plugging/unplugging, and screw-driving
- Robust transfer across five different objects with varying geometry and tolerances
- Resilience to spatial position disturbances across unseen instances
- Continuous policy blending without manual phase definitions or hard-coded thresholds
Why it matters
It allows industrial robots to perform complex, contact-rich tasks autonomously and securely without requiring expert manual tuning for every new scenario.
Abstract
As of today, automating contact-rich industrial manipulation processes, such as insertion, plugging, and screw- driving, is tedious and requires expert knowledge. The processes consist of programmable, common action units, like moving to a pose and establishing contact. However, the user still has to decide on fixed transition conditions to successfully complete each sub-action. Instead, we introduce a tactile memory-driven policy blending framework based on unified force-impedance control to enable autonomous transitions. At the core of our approach lies a structured representation of manipulation as a sequence of basic operations combined into relevant processes, each governed by real-time sensory feedback and annotated with process quality metrics (PQMs) that capture motion, force, and energy-level interactions. A bidirectional long-short-term memory (BiLSTM) model encodes recent PQM histories to determine the success of basic operations. Later, soft blending weights are generated, allowing smooth, adaptive transitions between operations without manual phase definition. To ensure functional safety during contact, we integrate an energy tank mechanism that enforces passivity by regulating energy exchange. The resulting control scheme enables robust and continuous tactile manipulation across variations in object geometry and spatial configurations. Experimental validation across four processes, five objects, and two position variants demonstrates successful transfer and resilience to position dis- turbances. Our findings highlight that learned tactile memory and quality feedback embedded in the control loop provide a principled foundation for intelligent, transferable manipulation, enabling fully autonomous process planning and execution in the future.