Media

Backgrounder

Artificial Intelligence in
Robotics

published by
International
Federation of Robotics
Frankfurt, Germany

May 2018
 IFR Background Information for
Journalists on Artificial Intelligence
in Robotics
1. What is artificial intelligence?

There is no agreed definition of artificial intelligence (AI) – or human intelligence. John

McCarthy, who coined the term artificial intelligence in 1956, complained, “As soon as it
works, no one calls it AI anymore”. The general characteristics of computer programmes
and methods currently termed AI include technologies that create and use a predictive
model to determine correlations within large datasets and, are able, as a result, to make
suggestions and execute transactions. Examples include: Recommendation engines;
spam filters; price engines; algorithmic trading systems; scheduling systems; medical
systems that process either images or records to generate diagnoses. The output of an
artificial intelligence algorithm can be used as the input to another programme or
physical machine which executes a task, such as a robot.

2. How is artificial intelligence applied in robots? What are the

benefits?

Artificial intelligence in robots gives companies new opportunities to increase productivity,

make work safer, and save people valuable time. Substantial research is being devoted
to using AI to expand robot functionality. Commercially available applications include the
use of AI to:
o Enable robots to sense and respond to their environment: This vastly increases the
range of functions robots can perform.
o Optimise robot and process performance, saving companies money.
o Enable robots to function as mobile, interactive information systems in numerous
settings from public spaces to hospitals to retail outlets, saving individuals time.

Examples
Sense-and-respond:
Identifying, picking, and passing objects: Traditionally, robots have been able to pick
up objects in a pre-programmed trajectory in which the object must be known and in the
expected place. Robots equipped with sensors can now be programmed using artificial
intelligence to identify specific objects regardless of their spatial location. 3D vision
software allows the robot to detect objects that are hidden by other objects. Through
machine learning, one of the technologies classed as AI, the robot can teach itself in a
very short time how to pick up an object it has not encountered before, applying the
appropriate level of force. The machine learning algorithm continues to improve as it
picks. Picking technology is advancing rapidly but it is currently very difficult for robots to
pick objects that are not rigid – for example, goods in plastic wrapping or floppy materials,

1 Artificial Intelligence Background Information for Media May 2018

or have irregular and variable shapes – such as fruit and vegetables – with an accuracy
and speed that is commercially viable.

Inspection: Artificial intelligence enables robots to inspect a wide variety of objects to

detect faults – from fruit and vegetables to underwater pipelines.

Mobility: AI technologies are enabling advanced mobility in robots. Whilst robots have
been mobile for over 60 years (the first Automated Guided Vehicle was introduced in
1953), AI enables robot mobility in unpredictable environments. Mobile robots have
traditionally been programmed to execute a specific set of manoeuvres in a linear
fashion, guided by signals (magnetic, laser, lidar) from devices installed for this purpose
in their environment. They have not traditionally been programmed to deal with
unexpected events – for example, if they encounter an obstacle, they can stop to avoid
collision, but they will not be able to find an alternative route to their goal. In contrast, an
AI-enabled mobile robot gets from A to B by building a real-time map (or updating a pre-
programmed map in real-time) of its environment and of its location within that
environment, planning a path to the programmed goal, sensing obstacles and re-planning
a path in-situ. Mobile robots using AI are in commercial use in a number of industries and
applications such as:
o Fetching and carrying goods in factories, warehouses, hospitals.
o Performing inventory management (mobile robots using RFID scanners or vision
technologies).
o Cleaning – from offices to large pieces of equipment such as ship hulls
o Exploration of environments dangerous for humans – e.g deep-sea, space,
contaminated environments.

Process Optimisation
AI is used to optimise robot accuracy and reliability. Most large industrial robot
manufacturers offer customers services using AI to analyse data from robots in real time
to predict whether and when a robot is likely to require maintenance, enabling
manufacturers to avoid costly machine downtime. Robot performance can also be
optimised through analysis of data from sensors - tracking, for example, its movement
and power consumption. The robot programme can be adjusted automatically based on
the output of the AI algorithm.

Predictive maintenance and process optimisation do not require AI. However, AI

technologies improve the speed and accuracy of both activities, resulting in cost savings.
In large-scale manufacturing automation projects robots are typically connected to other
machinery – including other robots – and AI is used to optimise the whole process,
analysing data from all machines.

Mobile Information Robots

Mobile robots are being used as information booths to assist customers in environments
such as hotels, hospitals, airports and shops. They can answer questions, lead
customers to requested products or locations and can video-link the customer to a
human service agent.

May 2018 Artificial Intelligence Background Information for Media 2

3. How will the use of AI in robots affect workers and jobs?

Intelligent robots can make work safer and more satisfying. Robots are assuming an
increasing range of jobs that are dangerous for humans, such as cleaning toxic or
infected environments.
AI expands the potential for robots to share tasks or processes with workers, taking on
those parts of the task or process that are unergonomic and repetitive, such as lifting,
fetching and carrying. These applications do not depend on AI, but AI technologies
enable the robot to work effectively in unpredictable or rapidly-changing environments.
See the IFR’s positioning papers on ‘The Impact of Robots on Productivity, Employment
and Jobs’ and ‘Robots and the Workplace of the Future’ (LINK) for more information on
how robots affect workers and jobs.

4. What is the focus of research and development in AI for robots?

The main areas of focus of AI research in robots are:

o Expanding picking capabilities to deal with objects that are not rigid or are not in static
locations.
o Expanding robot mobility to work effectively in non-standard environments (such as
rough terrain).
o Enabling control of robots through verbal commands and gestures.
o Making robots easier to programme: Robots can already be programmed by physical
demonstration (the robot is guided through the required motion path and force /torque
measurements are translated into code). Research is ongoing in applying AI to
enable robots to learn by watching video demonstrations, and by independent trial
and error. Reducing robot programming time and costs will increase robot adoption
by small-to-medium sized companies, making them more productive. Connected
robots will also be able to learn together, running effectively as parallel computers.

It is important to note that it will be some time before these technologies are commercially
viable, let alone ubiquitous - it typically takes several years or even decades for new
technologies to be adopted at scale 1.

1 The McKinsey Global Institute finds it can take between eight and 28 years for technologies
to be adopted at scale, from the point at which they become commercially available (A Future
That Works: Automation, Employment, and Productivity, McKinsey Global Institute, January
2017).

3 Artificial Intelligence Background Information for Media May 2018

5. Are existing regulatory systems and safety standards sufficient
to cover the risks of AI in robots?

The IFR believes that existing safety standards are sufficient to cover current
developments in the use of AI in robots 2 in commercial applications. Currently, no
additional regulation is required.

6. Is the idea of super-intelligent robots that humans are not able

to control a myth? Will robots outsmart us?

Fears of super-intelligent robots relate to General Artificial Intelligence (GAI). There is no

agreed definition of GAI, but broadly speaking, it the ability of a software programme to
apply concepts acquired in one specific context to multiple other contexts. In humans,
part of being able to apply learning from one context to another involves tacit
knowledge 3. Machines do not ‘learn’ like humans – they adjust parameters based on
goals and limits set by programmers with no understanding of what they are doing and
why.

GAI is not possible today, nor is it clear when, if ever, it might be. However, it is up to
humans to determine the parameters, or degree of autonomy, under which AI operates
as well as the level of unpredictability we are prepared to accept. AI programmes do
what they are programmed to do 4. The commercial relevance of GAI in robots is unclear.
The commercial viability of a robot is based on its ability to carry out tasks with high
precision and reliability, in compliance with strict safety and regulatory standards.

2 The ISO standards governing robot safety include ISO 10218-1 and ISO 10218-
2:2011 Safety requirements for industrial robots, Parts 1 and 2; ISO 13482:2014 Safety
requirements for personal care robots and ISO/TS 15066:2016 Safety requirements for
collaborative robots. In Europe, safety is regulated through the EU Machine Directive. In the
US, the safe application of robots is regulated by OSHA Guidelines for Robotic Safety.
3Even young children know, for example, that a giraffe cannot climb a ladder – something an
AI algorithm could only derive based on access to explicit information.
4 This is not to say that AI algorithms always do what the programmer intended. But
algorithms cannot define or repurpose their end goal. Concern has been raised about self-
learning algorithms that discover correlations between data without being tasked with finding
something specific or trained on existing examples. This can make it difficult for programmers
to understand why the algorithm has reached a particular conclusion. The IFR supports a
strong emphasis on ongoing research into algorithms that are able to explain their
computations to programmers and users.
May 2018 Artificial Intelligence Background Information for Media 4