Payload Quantification via Proprioceptive-only Sensing for a Single-legged Vertical Hopper

Abstract

Legged robots, in their various applications, can be sent to carry a payload during their locomotion. However, carrying a payload without knowing its weight would potentially impede the robot’s locomotion performance. This paper focuses on payload quantification for legged robots driven by quasi-direct drive brushless DC motors. A single-legged vertical hopper has been used for proof of concept. Experimental data on the ground reaction force were collected through proprioceptive-only sensing on the physical platform and compared to the predictions generated by an extended Spring-Loaded Inverted Pendulum model. The Bayesian method is then used for inferring the payload parameter within the model. It is found that the assumption of massless leg in developing this kind of reduced-order models for legged locomotion make them particularly inaccurate at the moment of impact and the leg compressing period after that. As a result, using data from the leg decompressing period of the stance phase makes the quantification results more useful. This shed light on future implementation of this framework in a real-time manner.