Fine Robotic Manipulation Without Force/Torque Sensor

Abstract

Force Sensing and Force Control are essential to many industrial applications. Typically, a 6-axis force/torque (F/T) sensor is installed between the robot's wrist and the end effector to measure the forces and torques exerted by the environment on the robot (the external wrench). While a typical 6-axis F/T sensor can offer highly accurate measurements, it is expensive and vulnerable to drift and external impacts. Existing methods aiming at estimating the external wrench using only the robot's internal signals are limited in scope. For instance, the estimation accuracy has mainly been validated in free-space motions and simple contacts, rather than tasks like assembly that require high-precision force control. In this letter, we present a Neural-Network-based solution to overcome these challenges. We offer a detailed discussion on model structure, training data categorization and collection, as well as fine-tuning strategies. These steps enable precise and reliable wrench estimations across a variety of scenarios. As an illustration, we demonstrate a pin insertion experiment with a 100-micron clearance and a hand-guiding experiment, both performed without external F/T sensors or joint torque sensors.