Ball-and-socket joint pose estimation using magnetic eld Tai Hoang1 Alona Kharchenko2 Simon Trendel2 Rafael Hostettler2

2025-04-27 0 0 3.92MB 14 页 10玖币

侵权投诉

Ball-and-socket joint pose estimation using

magnetic ﬁeld

Tai Hoang1, Alona Kharchenko2, Simon Trendel2, Rafael Hostettler2

1Technical University of Munich, Munich, Germany,

t.hoang@tum.de

2Devanthro – the Robody Company, Munich, Germany

Abstract. Roboy 3.0 is an open-source tendon-driven humanoid robot

that mimics the musculoskeletal system of the human body. Roboy 3.0

is being developed as a remote robotic body - or a robotic avatar - for

humans to achieve remote physical presence. Artiﬁcial muscles and ten-

dons allow it to closely resemble human morphology with 3-DoF neck,

shoulders and wrists. Roboy 3.0’s 3-DoF joints are implemented as ball-

and-socket joints. While industry provides a clear solution for 1-DoF joint

pose sensing, it is not the case for the ball-and-socket joint type. In this

paper we present a custom solution to estimate the pose of a ball-and-

socket joint. We embed an array of magnets into the ball and an array of

3D magnetic sensors into the socket. We then, based on the changes in the

magnetic ﬁeld as the joint rotates, are able to estimate the orientation of

the joint. We evaluate the performance of two neural network approaches

using the LSTM and Bayesian-ﬁlter like DVBF. Results show that in

order to achieve the same mean square error (MSE) DVBFs require sig-

niﬁcantly more time training and hyperparameter tuning compared to

LSTMs, while DVBF cope with sensor noise better. Both methods are

capable of real-time joint pose estimation at 37 Hz with MSE of around

0.03 rad for all three degrees of freedom combined. The LSTM model is

deployed and used for joint pose estimation of Roboy 3.0’s shoulder and

neck joints. The software implementation and PCB designs are open-

sourced under https://github.com/Roboy/ball and socket estimator

1 Introduction

Classical rigid robots, which often consist of a chain of rigid and heavy links, pose

a safety risk in unstructured environments. Soft robots, on the other hand, made

primarily of lightweight materials, have a great potential to operate safely in

any environment, especially during dynamic interaction with humans. Many soft

mechanical designs have been proposed in recent years. Musculoskeletal robot [5]

or more speciﬁcally, Roboy 3.0, an open-source humanoid robot, is one of the few

examples of the soft humanoid robot designs. It mimics the working principle

of the human musculo-skeletal system and morphology. Instead of having an

actuator directly in the joint like in classic rigid robots, Roboy 3.0 links are

actuated by a set of artiﬁcial muscles and tendons - series elastic actuators.

arXiv:2210.03984v1 [cs.RO] 8 Oct 2022

2 Tai Hoang et al.

Fig. 1: Tendon-driven hu-

manoid robot Roboy 3.0

Such design allows to passively store the en-

ergy in the muscles and makes the robot inher-

ently compliant. These properties, combined with

anthropomorphic morphology, turn to be bene-

ﬁcial in full-body motion mapping during tele-

operation as well collaborative object manipula-

tion and other close physical interactions with hu-

mans. However, achieving reliable, stable and pre-

cise joint and end-eﬀector-workspace control for

a tendon-driven humanoid [9] is challenging due

to diﬃculties in precise modelling of muscle co-

actuation and tendon friction among other things.

Thus, a high-accuracy robot state estimation is of

crucial importance to enable the implementation

of closed-loop control algorithms. There are two

types of joint being used in the upper body of

Roboy 3.0: 3-DoF and 1-DoF. The neck, shoul-

ders, and wrists are 3-DoF joint and elbows are

1-DoF. Determining the joint position for 1-DoF

joints is possible with high precision using oﬀ-the-

shelf sensors. However, the same solution is not

directly applicable for the 3-DoF joint case. We

therefore, in this paper provide a solution on both

hardware and software for this 3-DoF ball-socket

joint.

Our idea is based on using the magnetic ﬁeld

of strong neodymium magnets to determine the orientation of the joint. This has

long been a well known approach due to various reasons. First, it is a contact-

free measurement, and way more robust to the environment compared to other

contact-free sensors like optical [3] and vision-based [2]. Second, the measurement

value can be obtained at high rates with a high accuracy. The sensor devices are

also not expensive and have a suﬃciently small size which makes it easy to

integrate into our robot. However, obtaining the orientation directly from the

magnetic value in closed-form is challenging due to their non-linearity relation.

We therefore shift the focus on the data driven approach, or more particularly,

learning a neural network to do the transformation between the magnetic sensors

output and the joint’s orientation. Recently, several researchers have successfully

developed solutions based on this idea. Jungkuk [8] built a full pipeline to obtain

the orientation of a 2-DoF object with only a single sensor, Meier [11] developed

a system that can determine the human hand-gesture based on the magnetic

ﬁeld. Magnetic ﬁeld can also be used to determine the position of an indoor

robot, which is an active research area for indoor localization problems [12],

[14].

Ball-and-socket joint pose estimation using magnetic ﬁeld 3

The magnetic sensor used in this paper is a three-dimensional Hall-eﬀect

sensor. This particular type of sensor has a critical issue: it is very noisy and

since our ultimate goal is using the inferred orientation to close the control

loop, it is very important to mitigate this issue, otherwise our robot could not

operate safely and can easily damage itself and the environment. In this paper,

we proposed to use two types of neural network that can deal with this problem:

a recurrent-based neural network (LSTM) [1] and Deep Variational Bayes Filter

(DVBF) [6], [7]. We hypothesize that both of the approaches can smoothen

out the output signal. While the former does this based on the historical data,

the latter approach is designated to do this more explicitly with the integrated

Bayesian ﬁlter.

2 Hardware

(a) ball-in-socket joint (b) printed circuit board

Fig. 2: Hardware design of Roboy 3.0 shoulder ball-and-socket joint

Figure 2 shows the hardware design of the ball-socket joint of Roboy 3.0.

On the left image, the ball-in-socket includes two main components, the ball

and the socket. Inside the 3D-printed socket is the PCB with four 3D-magnetic

sensors TLE493d mounted around the circuit (Figure 2b). The sensor data is

automatically read out by the Terasic DE10-Nano FPGA dev kit using an I2C

switch TCA9546A. The sensors measure the magnetic ﬁeld strength generated

by three 10mm neodymium cube magnets mounted inside the ball. The magnets

文档加载中……请稍候！
如果长时间未打开，您也可以点击刷新试试。

下载文档到电脑，查找使用更方便

10 玖币 0人已下载

立即下载

摘要：

Ball-and-socketjointposeestimationusingmagneticeldTaiHoang1,AlonaKharchenko2,SimonTrendel2,RafaelHostettler21TechnicalUniversityofMunich,Munich,Germany,t.hoang@tum.de2Devanthro{theRobodyCompany,Munich,GermanyAbstract.Roboy3.0isanopen-sourcetendon-drivenhumanoidrobotthatmimicsthemusculoskeletalsyste...

展开>> 收起<<

Ball-and-socket joint pose estimation using magnetic eld Tai Hoang1 Alona Kharchenko2 Simon Trendel2 Rafael Hostettler2.pdf

共14页,预览3页

还剩页未读，继续阅读

声明：本站为文档C2C交易模式，即用户上传的文档直接被用户下载，本站只是中间服务平台，本站所有文档下载所得的收益归上传人(含作者)所有。玖贝云文库仅提供信息存储空间，仅对用户上传内容的表现方式做保护处理，对上载内容本身不做任何修改或编辑。若文档所含内容侵犯了您的版权或隐私，请立即通知玖贝云文库，我们立即给予删除！

Ball-and-socket joint pose estimation using magnetic eld Tai Hoang1 Alona Kharchenko2 Simon Trendel2 Rafael Hostettler2

相关推荐

开通VIP享超值会员特权

作者详情

相关内容

热门标签

举报选择: