IEEE Robotics & Automation Magazine - December 2019 - 32

and published on the use of humanoid robots in the aeronautical industry.
The main concerns are non-added-value tasks because they
include health risks, such as high-repetitive strain injury, that
require high precision and low dexterity. With Comanoid, we
targeted the tasks of pattern printing or bracket positioning on
the fuselage because they are made nearly everywhere and do
not require heavy payload nor dexterous manipulation. The
primary goal was to show the humanoid robot's ability to reach
the workspaces and for it to be precise with both the localization and positioning of the end effector in the structure.
In the framework of another research program, four use
cases of relative complexity were identified:
1) circuit breaker automation, which requires operation in
cluttered spaces
2) hydrofuge protection and cleaning, which requires managing equilibrium and in-line tool trajectory planning
3) torqueing automation, which requires multicontact balance
control in dynamic motion and force control
4) system installation, which requires manipulating flexible objects.
Other tasks, such as drilling and cockpit end-phase systems checks, were also suggested.
Constraints
Accessibility and manipulation needs are already very challenging. The additional constraints under which robots
operate render these challenges nearly unreachable, as
depicted in Figure 1.
● Any robot shall operate and share the same space with
human workers and be safe in all circumstances.
● Humanoid falling shall be demonstrated in worst-case
conditions on humans and even on airplane structures.
● The robot shall be able to access any spot starting from any
other spot and travel between the three working levels
using human means, i.e., without requiring changes to the
manufacturing infrastructure.
● No external sensors, such as external cameras, are allowed;
however, bringing additional lighting is permitted.
● Safety certification and compliance with existing internal
regulations and procedures will be established.
The good news is that workers follow strict procedures
and rules. The CAD model of any piece or part of the airplane

(a)

(b)

(c)

is available (including its exact shape and inertia parameters),
and a digital mockup is updated in parallel to the entire
assembly process.
Basic Technological Requirements
In light of previously established requirements and needs, we
have identified the following must-have technologies to be
developed further for the Comanoid project and implemented
in the HRP-4 and TORO humanoid robots by the end of the
project: 1) multicontact planning and control, 2) walking, 3)
embedded localization and mapping, 4) integrating visual and
force control into the task-space optimization control, and 5)
contact detection and safety.
Multicontact Planning and Control
Workers often operate in very unergonomic postures, as
shown in Figure 1. In such cases, they use multicontact postures to relax stress, improve their equilibrium, or cast their
body posture (e.g., contact with the knees, shoulders, and
back). Humanoids shall be endowed with similar multicontact behaviors, which is a key technology of Comanoid; this
allows for the transformation of a humanoid robot into a
reconfigurable multilimb system that can adapt to narrow
spaces and increase its equilibrium robustness and manipulation capabilities.
Multicontact Planning
A recent review of multicontact technology [1] reveals that this
problem is consensually approached through the use of a multilevel computation. The first level plans contacts around a sort
of free-motion preplanned "guide" that exploits some properties under different hypotheses (there are many variants) but
has as an output a set of contact sequences and associated transitions. In a second phase, the latter are inputs for a simplified
model [e.g., center of mass (CoM)] that generates a consistent
centroidal dynamics trajectory under balanced criteria (presented later in this article). In the last phase, this generated trajectory is the input to the whole-body controller, which also
deals with other task objectives and constraints. The problem
in such a multilevel computation is to make each phase potentially feasible for the upcoming one and ensure that, if any
phase turns out to be infeasible for the given input, the latter is
quickly recomputed from the current state. Although

(d)

(e)

Figure 1. Examples of the working areas and conditions in aircraft manufacturing sites. (a) The cargo area. (b) Kneeling and using a
small ladder to access the roof. (c) The cockpit area. (d) The removal of the shop floor to access in-between levels. (e) A CAD digital
model at a 1:1 scale of the A350 mockup used to demonstrate the technologies for most of these scenarios.

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