IEEE Robotics & Automation Magazine - December 2019 - 50

Gain (dB)

Joint Torque Control
It is desirable to have a torque-control system with high
bandwidth and low tracking error to produce extremely
compliant behavior and
precise movement. However, actuators with
Crawler-type and
torque sensors must be
compact for robots to
quadruped robots cannot
work in narrow environments. It is possible to
climb vertical ladders
use an encoder [12] as a
spring deflection sensor
and may have difficulty
to enable joint torque
control. However, these
mounting steep stairs.
have a limitation because they require a relatively large deflection to ensure sufficient torque-sensing
resolution. Thus, we employed a system based on strain

0
-10
-20
-30
-40

Hip Roll
Knee Pitch

gauges attached to flexure elements [13]. They are thin
films, have sufficient sensitivity, and are attached to relatively stiff sensors [14].
The actuator unit consists of a brushless dc motor, harmonic drive gear, and torque sensor. The design of our
torque-controlled actuator is characterized by a ringshaped torque sensor placed outside of the motor or the
gear [10]. This layout shortens the axial length of the actuators. Figure 4 shows the structures of some joints. Our
sensors are designed to have a thin, cylindrical flexure part
with strain gauges attached and flanges to connect gears
and links, as shown in Figure 4(d) and (e). The thickness
of the flexure part is determined as 0.15-0.2 mm to maximize the resolution while satisfying the strength requirements. Table 2 shows the specifications of the designed
torque sensors.
A proportional-derivative controller with a disturbance
observer for the torque-control system is chosen because of its
ease of tuning and its robustness. The time constant of the
feedback control is 1.5 ms. A Bode plot of the torque-tracking
performance is shown in Figure 5. Consequently, a control
bandwidth of over 100 Hz is achieved. The robot was hanging
above the ground during these measurements, so the joint
loads were not fixed.

Phase (°)

System Overview
E2-DR is equipped with an electronic control unit
(ECU) for managing its motions (the motion-control
ECU), with an Intel Core i7 2.4-GHz CPU. There is
also an ECU for perception and recognition (the recognition ECU), with an Intel Core i7 2.2-GHz CPU
2
3
10
10
and a Haswell graphics processing unit. In addition,
Frequency (Hz)
the robot is equipped with distributed motor control
ECUs and drivers for driving motors, joint torque senFigure 5. A Bode diagram of the closed-loop torque tracking
sors, and six-axis force torque sensors attached on its
for the hip roll and knee pitch joints, showing the frequency
response from the desired torque to the measured torque.
wrists and ankles, as shown in Figure 6. The motioncontrol ECU communicates with
those devices through its internal
network, with a sampling rate of
1 kHz. The robot is remotely teleSensor ECU
WLAN
LRF
TOF
operated by communicating with a
Joint Torque
Sensor
console PC operated via a wireless
local area network.
Internal Network
To reduce the risk of cables disconCommunication
Recognition Ether
MotionMotor
Driver
necting because of a high harness
ECU
Control ECU
occupancy rate, we use 0.5-mm-diameter optical fiber for internal network
Serial
Power-Supply
Hand-Brake
communication. The major concern
CameraBMU
Control
Control
Control Unit
for optical fiber when used in a robot's
ECU
Unit
joints is mechanical tolerance and
Sensor ECU
IMU
Force Torque
attenuation during twisting and bendCamera
Sensor
LED
ing. After performing cycle endurance
tests of twisting and bending optical
fiber 1 million times, it was concluded
Figure 6. The electrical system and internal network configuration of E2-DR. WLAN:
that the cable meets the tolerance and
wireless local area network; IMU: inertial measurement units; BMU: battery
attenuation requirements.
management unit.
50

0
-100
-200
-300
-400

IEEE ROBOTICS & AUTOMATION MAGAZINE

DECEMBER 2019

IEEE Robotics & Automation Magazine - December 2019

Table of Contents for the Digital Edition of IEEE Robotics & Automation Magazine - December 2019