IEEE Robotics & Automation Magazine - December 2019 - 38

assess new technological ideas in production and assembly
operations; however, a significant number of parts were
machined from metal blocks instead of being assembled from
production parts. Given that the A350 has a significant number of composite parts, the materials and textures might differ
from the aircraft even if the partial geometry is very close.
Demonstrator of the HRP-4 Humanoid Robot
The demonstrator of the HRP-4 humanoid robot has the following variants w.r.t. the previously described scenario:
● Localization everywhere uses SLAM solely.
● Climbing/walking uses its own developed controller.
● Stair climbing is accomplished without the use of handrails.
● Task-aware, multicontact planning uses the left hand for
additional support to lean toward the structure.
● Localization (also grasping) of the bracket and the taskaware contact uses visual servoing.
● The bracket is glued and released with force control.
● Safety achieved using whole-body capacitive sensing is
demonstrated interactively after the robot exits the working area.
● All of the demonstrator's tasks are completed autonomously by the robot, but the operator validated each step prior to
transitioning to the next (this can be skipped) and intervened if SLAM relocalization failed.

Two full trials are performed without securing the robot.
HRP-4 did not return to floor 0.
Figure 7 illustrates the main components of the task specification and control architecture for control of the HRP-4
humanoid robot, using the technological bricks described in the
"Basic Technological Requirements" section. This architecture is
currently implemented in several other robots, such as the SoftBank Robots Pepper and Nao, other HRP robots, and the Sawyer robot. It has three main components:
1) low-level and high-performance C++ libraries for robotic experts
2) a unified controller interface mc_rtc, which is the control
framework used to facilitate the development of controllers
and the integration of new robots
3) simulation/control interfaces that are "glue layers"
between mc_rtc and simulation, e.g., virtual robot experimentation platform (V-REP), Choreonoid, or robothardware interface.
The mc_rtc framework is written in C++ but permits the
writing of any robot controller in either C++ or Python. Our
low-level libraries are mainly concerned with the mathematical and numerical aspects of control, i.e., computing all of the
required matrices and vectors correctly in a timely manner
and setting up and solving optimization problems. The mc_
rtc framework brings simpler interfaces, simpler semantics
●
●

High-Performance C++ Libraries
Tasks

Spatial Vector Algebra
C++11 Implementation

Rigid Body Dynamics
Algorithms Implementation

RT QP-Based Control

Spatial Quantities,
Representations, and Operations

Rigid Body Representation and
Dynamics Algorithms

QP Solver, Low-Level Tasks, and
Constraints System

Unified Controller Interface (mc_rtc)
Single Interface for Integration With Simulation/Robots
Unified Controller Interface to Build New Controllers (Python/C++)

ROS

Extensive Tooling to Program the Complex Controller
High-Performance Logging and Log-Analysis Tools

Integration

Server-Based GUI
Statecharts Implementation
Command
(Desired Joint Position/Torque)

Sensor Data
(Encoders, Forces, IMUs ...)

Simulation/Control Interface
Light Software Layer Above Existing Simulation Softwares/Control Framework
OpenRTM

ROS

HRP Robots and Choreonoid

Visualization, Gazebo, and Control

v-rep

V-REP

Dynamics Simulation

Figure 7. The control architecture of HRP-4. GUI: graphical user interface; IMU: inertial measurement unit.

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