A Learning-Based Approach for Estimating Inertial Properties of Unknown Objects from Encoder Discrepancies

TL;DR

This letter proposes a learning-based approach for estimating the mass and center of mass (COM) of unknown objects without using force/torque sensors at the end effector or on the joints, and has been demonstrated on a 4-degrees-of-freedom robot arm.

Abstract

Many robots utilize commercial force/torque sensors to identify inertial properties of unknown objects. However, such sensors can be difficult to apply to small-sized robots due to their weight, size, and cost. In this paper, we propose a learning-based approach for estimating the mass and center of mass (COM) of unknown objects without using force/torque sensors at the end-effector or on the joints. In our method, a robot arm carries an unknown object as it moves through multiple discrete configurations. Measurements are collected when the robot reaches each discrete configuration and stops. A neural network is designed to estimate joint torques from encoder discrepancies. Given multiple samples, we derive the closed-form relation between joint torques and the object's inertial properties. Based on the derivation, the mass and COM of object are identified by weighted least squares. In order to improve the accuracy of inferred inertial properties, an attention model is designed to generate weights of joints, which indicate the relative importance for each joint. Our framework requires only encoder measurements without using any force/torque sensors, but still maintains accurate estimation capability. The proposed approach has been demonstrated on a 4 degree of freedom (DOF) robot arm.

Figures (6)

• Figure 1: Block diagram of the proposed approach. (a) Training process. The training data are collected with known objects. The torque model and attention model are trained sequentially. (b) Testing process. The testing data are collected with unknown objects. The torque model estimates the joint torque, and the attention model generates a weight matrix. The mass and COM are solved by weighted least squares.
• Figure 2: (a) Schematic of an $N$-DOF robot carrying an object. (b) Free body diagram of the $j$-th link.
• Figure 3: The architectures of proposed neural networks. The first row illustrates the torque model and the attention model. The second row illustrates submodules in the above models.
• Figure 4: (a) OpenMANIPULATOR-X robot manipulator. (b) Experimental setup. (c) Training objects. (d) Testing objects.
• Figure 5: (a) Snapshots of robot carrying objects. (b) Results of vertical force estimation in continuous experiments.