DULA & DEBA: Differentiable Ergonomics Models for Postural Optimization

Abstract

Ergonomics and human comfort are essential concerns in physical human-robot interaction applications. Defining an accurate and easy-to-use ergonomic assessment model stands as an important step in providing feedback for postural correction to improve operator health and comfort. In order to enable efficient computation, previously proposed automated ergonomic assessment and correction tools make approximations or simplifications to gold-standard assessment tools used by ergonomists in practice.

In order to retain assessment quality, while improving computational considerations, we introduce DULA, a differentiable and continuous ergonomics model learned to replicate the popular and scientifically validated RULA assessment. We show that DULA provides assessment comparable to RULA while providing computational benefits. We highlight DULA's strength in a demonstration of gradient-based postural optimization for a simulated teleoperation task.

Motivation

Autonomous Postural Optimization has received substantial attention in research with the new technologies around humans such as collaborative robots, smart personal trainers, and VR systems. In these systems, the interacting agent should consider human comfort and ergonomics in its behaviour and motion planning.

Developing a model of human comfort lies at the heart of effective postural optimization. p-HRI researchers have proposed several computational models for assessing ergonomics and human comfort in terms of peripersonal space, muscle fatigue, and joint overloading. In contrast ergonomists have provided simpler models which are easier for human experts to calculate by hand and as such are more common in practice. Importantly these models are supported by extensive human subject studies that validate their effectiveness on reducing ergonomic risk factors.

Among all risk assessment tools, RULA and REBA depend most on human posture and provide quantitative scores, making them good choices for postural optimization applications. However, the discrete scores and the presence of plateaus in RULA and REBA create challenges when using them in gradient-based postural optimization. Based on our experience, using the risk assessment models directly in gradient-free optimization is time-expensive and the plateaus often prevent progress toward the global optimal solution in postural optimization. Thus, researchers in pHRI often use approximations of ergonomic assessment models in gradient-based postural optimizations; quadratic approximations being the standard approach in the literature. However, these approximations deviate far from the scientifically validated assessments, causing doubt that they can reliably provide the same level of ergonomic benefit.

Our Solution

To overcome these issues, we introduce the Differentiable Upper Limb Assessment (DULA), and Differentiable Entire Body Assessment (DEBA), differentiable and continuous risk assessment models that are learned using a neural networks to replicate the popular RULA and REBA survey tools. Instead of discrete scores DULA and DEBA report the risk score as a continuous real numbers. Furthermore, they provides the gradient of the risk with respect to each joint enabling efficient use in postural optimization.

Method

To build a differentiable models, we developed a dataset of 7.5 million upper body and full body postures of a human model. We label the upper body and full bofy postures dataset using RULA and REBA scores, respectively.

Moreover, to learn a continuous and differentiable function for RULA and REBA, we designed fully-connected regression-based neural networks. While RULA and REBA provide discrete integer scores from 1 to 7 and 1 to 12, respectively, we choose to predict continuous labels. If we had instead performed multi-class classification based on the discrete labels the resulting model would be less useful in optimization as there is no natural choice of smooth objective to minimize or constrain ergonomic cost. This would negate our desire for a computationally useful model. Hence, we perform regression to the multi-class labels.

Results

Testing accuracy for DULA: 99.73% !

Related Publications:

1. "Ergonomically Intelligent Physical Human-Robot Interaction: Postural Estimation, Assessment, and Optimization", Amir Yazdani, Roya Sabbagh Novin, Andrew Merryweather, Tucker Hermans, submitted to AAAI AI-HRI, 2021. arXiv

2. "DULA: A Differentiable Ergonomics Model for Postural Optimization in Physical HRI", Amir Yazdani, Roya Sabbagh Novin, Andrew Merryweather, Tucker Hermans, RSS workshop Robotics for People, 2021, arXiv

3. "Posture Estimation and Optimization in Ergonomically Intelligent Teleoperation Systems", Amir Yazdani, Roya Sabbagh Novin, Companion of the ACM/IEEE HRI, 2021, PDF