07

ROBOTICS

KEY TAKEAWAYS Overview

° Although robots today are mostly used for the What is a robot? Researchers do not universally agree
Three Ds (dull, dirty, or dangerous tasks), in the on the definition of a robot, but a consensus seems
future they could be used for almost any task to have emerged that at the very least a robot is a
involving physical presence, because of recent human-made physical entity with ways of sensing
advances in AI, decreasing costs of mobile com- itself or the world around it and ways of creating
ponent technologies (e.g., cameras in smart- physical effects on that world.1
phones), and designs enabled by new materials
and structures. The global robotics market is estimated at $25 bil-
lion today and poised for strong growth over the
° Robotics has and will transform many industries next decade. While projections vary greatly, some
through elimination, modification, or creation of consultancies estimate that the global market could
jobs and functions. be worth between $160 billion and $260 billion by
2030.2 The adoption of professional service robots
° Understanding and communicating how robots is expected to drive this growth. These types of
will affect people’s lives directly in their physical robots currently occupy a small sliver of the market,
spaces (e.g., security robots in malls) as well as but as technology continues to improve, they will
more existentially (e.g., transitioning jobs like find greater adoption in industries like medicine,
truck driving from human-driven to autonomous agriculture, and construction. The United States is
vehicles) will shape how the United States accepts projected to maintain a plurality of global robotics
and benefits from robotic technologies. revenues for the next few years.

85
 Importantly, robots must integrate many different and the robot’s own configuration. Such inputs
component technologies to combine perception inform decisions about what the robot should be
of the environment with action on the environment. doing in the next moment in time.
Perception requires generating a representation of
the robot’s environment and interaction with it. Action ° Control systems These components decide
requires the robot to make physical changes to itself how actuators should move based on readings
or the environment based on those perceptions. from sensors.

The key engineering challenges in robotics involve ° Materials Constructed of rigid materials and
the design of components (e.g., enabling visual or with joints based on ordinary bearings, traditional
other perception) and then integrating them within robots interact with their operating environments
a physical or mechanical structure to perform the in highly prescribed and structured ways. “Soft”
robot’s intended tasks (e.g., using perception to robots that are flexible and conform to the envi-
guide motions and actions) in different settings in ronment can offer better performance in the
a given environment. The physical structure could more unstructured and chaotic environments that
be regarded as a robot’s body. Further, different characterize most of the world, but the construc-
types of robots operate in different environments tion of soft robots often entails the creation of
(e.g., factories, homes, and even space), and each new materials or structures.
environment raises distinct complexities beyond
just technical performance. For example, working ° Power sources Tethered robots can be ener-
alongside humans raises critical issues of safety and gized from a power source on the “mother
liability. ship” indefinitely, while untethered robots need
self-contained power sources or sources that har-
The dependence of robotics on many different com- vest energy from the environment. A common
ponent technologies and nontechnological consid- portable power source is batteries, which drain
erations has an important practical consequence—it themselves quickly—too quickly for many practi-
takes a huge interdisciplinary effort, not just from cal applications of robots.
technologists but also from experts in other fields,
to move from a working prototype in a research lab ° Real-time programming As physical devices,
to a useful functional robot in the market. robots operate in real time and their compo-
nents must operate within the boundaries of
Important component technologies include: physical timelines determined by the operation
of the robot. An actuator that moves a tenth of
° Actuators These components enable move- a second too early may cause a robot hand to
ment (e.g., motors, grasping appendages). Today, fail at grasping an object. Deviations in timing
mechanical actuators are typically rigid, restrict- may have nothing to do with the programming
ing the environments where robots can operate. of the real-time microprocessor, but rather occur
These inflexible actuators make operating in con- because another subsystem in the robot failed
fined spaces and on irregular terrain or performing to operate on time or because something unex-
dexterous movements in unknown environments pected happened in the robot’s environment.
challenging.
While some robots use computer vision and other
° Sensors These receive real-time input about types of AI for understanding their environments
the immediate physical environment of the robot and decision making, robotics and AI do not always

86 STANFORD EMERGING TECHNOLOGY REVIEW

 FIGURE 7.1 Not all robots use artificial intelligence

Key Developments
Robots are used across a wide range of sectors in a
variety of ways. Prominent examples include:
Artificial
Robotics Intelligence ° Manufacturing Many assembly lines use station-
ary robots (see figure 7.2) to undertake repetitive
tasks such as welding at high speeds and with great
precision.5 The environment around such robots is
highly structured and carefully controlled to mini-
mize the need for robot cognizance of surround-
Artificially ings. In most cases, these robots work in isolated
Intelligent cells without any physical interaction with humans.
Robots

° Warehouse logistics Autonomous robots bring

merchandise stored in very large warehouses to a
central point for packaging.
go together (see figure 7.1). Some robots, for exam-
ple, are tele-operated, requiring a human opera- ° Surgery Mostly tele-operated today, surgical
tor to control or direct most aspects of the robot’s robots assist with minimally invasive surgery. Sur-
behavior.3 Other robots can do a few things autono- geons can reduce the size of the incisions that are
mously without AI, such as maintaining themselves needed for treatments and thus reduce surgical
in a fixed position or entering a “cruise control” risks.6 A surgical robot typically includes a camera
mode, like the early versions of the Roomba, a arm and surgical instruments attached to mechan-
series of robotic vacuum cleaners introduced in ical arms controlled by a surgeon at a console op-
2002. There are robots that rely on AI for extensive erating the robot.
capabilities for autonomous behavior and decision
making, such as the self-driving taxis approved for
business in San Francisco in 2023. Robots with vary- FIGURE 7.2 Automotive assembly line robots
ing degrees of autonomy have been used in every-
thing from delicate surgical procedures to space
exploration.

Robots are used primarily for the Three Ds: human

work that is dull, dirty, or dangerous. Human attention
to a task waxes and wanes, whereas robots do not
get bored. Human life and well-being are precious
compared to the physical damage that a robot might
experience in doing a task. And robots can survive
in much more hazardous or extreme environments—
cleaning up nuclear reactors and exploring Mars, for
example—than humans can.4 Source: Carol M. Highsmith Collection, Library of Congress

07 Robotics 87
 FIGURE 7.3 Mars Curiosity rover

Source: NASA

FIGURE 7.4 Sentry autonomous underwater vehicle,

Woods Hole Oceanographic Institution FIGURE 7.5 Search-and-rescue robot

Source: NOAA Ocean Exploration Source: National Institute for Standards and Technology

° Science and exploration Robots have been underneath rubble, bringing communications,
used to explore other planets,7 the vast littoral oxygen, food, or medicine (see figure 7.5).13
ocean zones,8 buildings,9 the insect world,10 and
the human body.11 Planetary rovers (see figure 7.3) ° Security Mobile robots in parking lots and
and remotely operated underwater vehicles (see buildings such as shopping malls provide tele­
figure 7.4) are two examples of such robots. presence for centrally located security personnel.

° Food production Agricultural robots12 can ° Military services Robots have been developed
help harvest crops by picking fruit and maintain that help to perform a variety of military services,
farmland by weeding. Drones can inexpensively including load transport, surveillance and recon-
survey farmland, and robot-operated green- naissance, mine clearance, and armed sentry duty.
houses enable food production.
° Transportation Autonomous vehicles and trucks
° Disaster assistance robots These robots can are the most common examples.14
maneuver in collapsed spaces to reach victims

88 STANFORD EMERGING TECHNOLOGY REVIEW

Robots excel at working in structured environments components, and novel designs enabled by new
like manufacturing plants where conditions are pre- materials. Researchers in robotics today15 are work-
dictable. Their precision, speed, and tirelessness ing in areas such as:
allow robots to surpass human performance in repet-
itive tasks on the assembly line. However, humans Haptic technology for robots Sensations of
still outperform robots in unstructured or poorly touch and feel are useful in manipulation and
structured environments. Constraints on power, fuel many social interactions—for example, touch feed-
supplies (e.g., battery life), and sensors also limit the back is vital to a surgeon when palpating tissue.
ability of robots to compete with humans. Haptic technology enables doctors to use touch
as an input device to robotic surgical systems and
Most important, human intelligence and adaptabil- tele-operated robots, facilitating more intuitive
ity give people a major advantage in chaotic, real- control and stronger physical connections with
world environments. For example, robots struggle remote environments.16
to navigate cluttered spaces or manipulate unfamil-
iar objects. Even advanced robots cannot yet match Robotic movement through self-deformation Soft
humans’ intuitive understanding of physics and the robots are safer for humans—a human–soft robot col-
ability to improvise solutions. Robots rely heavily on lision creates an impact between soft human tissue
precise planning and control, so encountering small, and a soft robot body and is less likely to result in
unexpected changes may cause catastrophic failure. injury. Soft robots’ ability to deform also affords novel
Humans, on the other hand, can apply their past methods of locomotion and manipulation. Soft robot
experiences and reasoning to adapt to unfamiliar technology can be applied in haptic interfaces, search
situations on the fly. and rescue, and medicine.17 Soft robots can also use
materials that are “continuum, configurable, and
adaptable with functionalities relying on high degrees
of freedom shape morphing.”18

Over the Horizon Robot design New types of robots may be needed
for operation in difficult-to-access or uncertain envi-
New Robotic Technologies
ronments. For example, robots for space explora-
A growing direction in robotics is the one where tion include small free-flying robots to operate with
robots and humans work together to capitalize on space stations, robot manipulator systems for use
the advantages of each. on craft that orbit planets, rovers that can jump in
low-gravity environments, and robots that can move
Advances in artificial intelligence and soft robotics by stepping with adhesive pads.19
may help robots become more capable in unstruc-
tured spaces. Machine-learning algorithms could Wearable robots A wearable robot such as an
enable robots to perform well in environments that exoskeleton enhances human mobility during activ-
they have never seen before or never been pro- ities like walking and running and may assist people
grammed to encounter. Flexible, nonrigid robots with physical impairments.20 They can also be used
can perform in a variety of environments, even those to augment human muscle power (see figure 7.6),
inaccessible by humans. thereby enabling people to lift large loads.

Advances in robotics will be linked to advances in Robotic manipulation Situations in which grasping
artificial intelligence, the decreasing cost of mobile an object is necessary often do not provide perfect

07 Robotics 89
 FIGURE 7.6 Sarcos Guardian XO Max wearable Artificial intelligence for robotics This refers to
exoskeleton the challenge of giving robots the ability to learn
how to learn and exhibit commonsense, intelligent
behaviors.

Human-robot interaction (HRI) HRI focuses on

understanding, designing, and evaluating robots
for use by or with humans.24 Humans and robots
working closely together, for example, may have
unpredicted, unintentional physical contact that
may disrupt robot operation and cause safety issues
for the human.25 Another challenge is the design of
these interactions in ways that accommodate social
norms, are natural and seamless, and allow robots
to exhibit more familiar and comfortable behaviors
to humans.

Challenges of Innovation
Source: Sarcos Technology and Robotics Corporation
and Implementation
Supply chain issues are one of the most important
near-term infrastructure challenges in robotics. The
information about the object, the grasping device, robotics field involves the integration of multiple
and the relative positions of everything in the envi- foundational technologies, which means progress is
ronment. So it’s important to develop manipulation heavily reliant on global supply chains for parts such
sequences that work with uncertainty.21 as chips and materials. The more far-flung and com-
plex the international innovation supply chain, the
Bio-inspired robots These are designed based on more slowly innovation will move when disruptions
fundamental biological principles and could include to the supply chain occur.
living components.22
To illustrate, DJI is a Chinese company that con-
Robot swarms These small, modular robotic com- trols a large share of the airborne drone market.
ponents act in coordination as a team to perform One important reason for this dominance is the
tasks.23 entire supply chain for DJI drones is self-contained

Advances in robotics will be linked to advances in

artificial intelligence, the decreasing cost of mobile
components, and novel designs enabled by new materials.

90 STANFORD EMERGING TECHNOLOGY REVIEW

within one region of China. Start-up companies in traction in the economy. If self-driving trucks come
the United States working in this space are generally to dominate the landscape, we might see positions
forced to turn to Chinese suppliers as the US supply in logistics and fleet management increase, although
chain for drones is fragmented, making it cumber- these kinds of jobs entail a different skill set than that
some, expensive, and slow to deliver. required for driving trucks.

Policy, Legal, and Regulatory Issues ACCOUNTABILITY, REGULATION, AND LIABILITY

Societies routinely hold people accountable for

ROBOTS AND THE FUTURE OF WORK
harmful actions. People who cause accidents that
Robots have enormous potential to affect jobs harm people or property face liability for that harm.
involving physical labor and presence in much the Soldiers in war who kill civilians because they use
same way that AI might affect jobs involving knowl- their weapons indiscriminately are guilty of war
edge and expertise. When realized in the market- crimes. But as robots assume roles that call for simi-
place, robotics is likely to eliminate some job types, lar decision making, how should concepts of individ-
create new job types, and modify the responsibilities ual accountability evolve?
and duties of jobs that remain.
Some questions include:
Consider truck driving, one of the few well-paying
jobs available to Americans without a college ° What parties should be held accountable for
degree. The profession is likely to experience the harm occurring when robots are involved and
following: how should those determinations be made?

Job elimination Long-haul truck driving is likely ° How and to what extent, if any, do robots involved
to be one of the first jobs eliminated when auton- in incidents that harm people or destroy property
omous land vehicles become feasible because disproportionately or improperly attract liability
fuel costs can be significantly reduced when trucks lawsuits?
drive close to one another on highways—possible
only with machine-speed reaction times rather than ° How can existing regulatory regimes for trans-
human reaction times. portation safety and medical safety, for example,
keep pace with evolving robotics technology?
Job modification Many truck-driving jobs entail
both long-haul and local driving. Automation may ° How and to what extent, if any, are lethal autono-
take on more responsibility for long-haul driving, mous weapons ethically and morally permissible?
but navigating local driving conditions in a far less
structured environment will take much longer to ° How is the safety of robot operation best assured?
automate, which means truck-driving jobs may How, if at all, should safety trade off against other
become more local in nature. But whether more performance objectives?
local truck-driving jobs will continue to offer com-
parable compensation to those jobs of today is not ° What are the appropriate standards of perfor-
at all clear. mance for robots? Robots often perform tasks that
humans also perform. Should the standard be that
Job creation Robot repair and technician jobs the robot does the task nearly perfectly? Or is it
are likely to be a major job category as robots gain adequate to perform it better than a human being?

07 Robotics 91
 If the latter, should the reference human being be jobs.26 It has been suggested that robots could
an average person or a person who performs the help the elderly care for themselves by providing
task much better than most other humans? emotional support or cognitive therapy; enabling
remote access for doctors and nurses; and entertain-
ing home dwellers, monitoring them for falls, and
SOCIETAL ACCEPTANCE
helping them with housekeeping, lifting, and bath-
For robots to be widely used throughout society, ing needs. By assuming part of the eldercare labor
their presence and operations should not cause force, robots could allow a limited number of care
human discomfort, unease, or fear. Experience sug- workers to do their jobs more efficiently and easily.
gests that many humans are disturbed by robots that
look like humans (see figure 7.7), such as in health- However, in one study of robots for eldercare in a
care settings. nursing home, several challenges emerged, includ-
ing the imposition of additional work burdens on
Some advocate the use of robots for eldercare, sug- human caregivers, the need for close monitoring
gesting that an aging population will create demands of the robots, and the displacement of social and
for services that cannot be met by future pools of communication-oriented tasks that reduced oppor-
workers interested in and qualified to perform those tunities for human connection.27

FIGURE 7.7 A very human-looking robot

Source: International Telecommunication Union, via Wikimedia Commons, CC BY 2.0

92 STANFORD EMERGING TECHNOLOGY REVIEW

ROBOETHICS IMPORTANCE OF MULTIDISCIPLINARY CENTERS

Roboethics addresses what ethical guidance should Robotics is a multidisciplinary field. Key develop-
be programmed into robots so they do not behave ments will occur not only in labs that are studying
in ways that humans regard as unethical or lead to “robotics” but also in those studying the underlying
unethical outcomes. A related definition suggests technologies that make robotics possible. These key
that roboethics refers not to the ethics of robots but fields include materials science, computer science,
rather to the human ethics of robot designers, man- and artificial intelligence. For example, if new mate-
ufacturers, and users.28 rials are developed by a materials science research
group, they could increase the capabilities of a
One particularly clear example of roboethics arises robotic gripper by allowing it to be equally strong
in a military context of lethal autonomous weap- but with less weight. Algorithms designed by com-
ons systems that can select a target and then act puter scientists could greatly improve the predictive
to destroy it without human intervention. Given power used by autonomous vehicles.
military pressures to act more rapidly on the bat-
tlefield, many observers believe that such decisions Multidisciplinary centers that bring together con-
will inevitably be turned over to computers. Others tributing research labs can facilitate cooperation.
recoil at the prospect, arguing it is ethically wrong to Individual labs can be focused on specific areas and
make decisions in matters of life and death without perform their research in the context of an overall
human input. robotics framework; the center itself can take the
role of integrating technologies to see how they fit
But ethical issues also arise in more benign con- together.
texts. For example, what are the ethics of large-scale
deployments of eldercare robots? Granting that Robotics research in academia is often coupled
many of the issues faced by an aging population closely to industrial counterparts. The confluence is
relate to physical needs that either robots or home natural because research in robotics focuses so heav-
health caregivers could meet, what of their emo- ily on applications—and applications are the focus
tional needs? How and to what extent, if any, can of industrial attention. In some subareas of robot-
interactions with an artificially intelligent eldercare ics research, such as human-robot interactions, the
robot provide comfort, empathy, and compassion sophistication of industrial activity outstrips that of
for the person under its care? academia. In some other areas, academic research
pushes in directions that industry is not pursuing,
such as theoretical aspects of robotics.
PRIVACY AND SECURITY

Robots—especially mobile robots—often raise pri- Multidisciplinary centers like the Stanford Robotics
vacy and security concerns in unexpected ways. Center (SRC) are also useful for providing points of
Cameras mounted on drones appear to raise many engagement with industry, pairing private compa-
more privacy concerns than stationary cameras. nies with researchers on campus who have similar
Drones have also recently demonstrated the ability interests. In return for funding, corporate partners
to open closed doors and therefore fly into areas can more easily access new research and future
that were once guaranteed to be drone-free. employees.

07 Robotics 93
 RESEARCH FUNDING NOTES
Funding for robotics research comes from a vari-
1. Ralph Lässig et al., “Robotics Outlook 2030: How Intelligence
ety of sources. The National Science Foundation and Mobility Will Shape the Future,” Boston Consulting Group,
supports foundational robotics research. Medical June 28, 2021, https://www.bcg.com/publications/2021/how
-intelligence-and-mobility-will-shape-the-future-of-the-robotics
robots can be funded by the National Institutes of -industry.
Health. Funding by the Defense Advanced Research 2. Statista, “Robotics—Worldwide,” accessed August 30, 2023,
Projects Agency (DARPA) is also common, with the https://www.statista.com/outlook/tmo/robotics/worldwide.
Department of Defense sponsoring robot compe- 3. National Research Council, Virtual Reality: Scientific and Tech-
nological Challenges (Washington, DC: The National Academies
titions to drive innovations they believe could be Press, 1995), 313, https://doi.org/10.17226/4761.
applicable to their use cases. 4. Alan Winfield, Robotics: A Very Short Introduction (Oxford:
Oxford University Press, 2012), 31, https://doi.org/10.1093
/actrade/9780199695980.001.0001.
Another important source of research funding is
5. Winfield, Robotics, 20–22.
the private sector. Tech companies like Google and
6. Winfield, Robotics, 34–35.
Amazon want to know what is coming down the 7. Mars Exploration Rovers, “Rover Update,” National Aeronau-
pipeline regarding robotics. They invest in academic tics and Space Administration, accessed August 30, 2023, https://
labs and contribute to affiliate programs like those mars.nasa.gov/mer.
8. Bjorn Carey, “Maiden Voyage of Stanford’s Humanoid Robotic
of Stanford’s SystemX Alliance and the Stanford Diver Recovers Treasures from King Louis XIV’s Wrecked Flag-
Robotics Center. ship,” Stanford News, April 27, 2016, https://news.stanford
.edu/2016/04/27/robotic-diver-recovers-treasures.
9. “A Research Group in Carnegie Mellon University’s Robotics
The amount of industrial funding devoted to robot- Institute Is Creating the Next Generation of Explorers—Robots,”
ics research and development far outstrips the Science Daily, July 19, 2023, https://www.sciencedaily.com
amount available to academic researchers. Partly /releases/2023/07/230719145936.htm.
10. National Science Foundation, “Award Abstract #1941933:
as a consequence, robotics faculty are being lured CAREER: Fast, Furious and Fantastic Beasts: Integrative Principles,
away from academia with compensation packages Biomechanics, and Physical Limits of Impulsive Motion in Ultra-
large enough to warrant concern at universities.29 fast Organisms,” accessed August 30, 2023, https://www.nsf.gov
/awardsearch/showAward?AWD_ID=1941933.
11. Jacob Biba, “Microrobotics: Tiny Robots and Their Many
Uses,” Built In, September 6, 2022, https://builtin.com/robotics
/microrobotics.
12. Stephen Gossett, “16 Agricultural Robots and Farm Robots
You Should Know,” Built In, March 7, 2023, https://builtin.com
/robotics/farming-agricultural-robots.
13. “Robots in Disaster Response: Helping First Responders in
Emergency,” RoboticsBiz, June 16, 2023, https://roboticsbiz.com
/robots-in-disaster-response-helping-first-responders-in-emergency.
14. Center for Sustainable Systems, “Autonomous Vehicles
Factsheet,” University of Michigan, September 2022, https://css
.umich.edu/publications/facsheets/mobility/autonomous-vehicles
-factsheet.
15. Guang-Zhong Yang et al., “The Grand Challenges of Science
Robotics,” Science Robotics 3, no. 14 (January 2018), https://doi.org
/10.1126/scirobotics.aar7650.
16. “CHARM,” Collaborative Haptics and Robotics in Medicine
Lab, Stanford University, accessed August 30, 2023, https://charm
.stanford.edu.
17. Alexander M. Kübler et al., “A Multi-Segment, Soft Growing
Robot with Selective Steering,” 2023 IEEE International Con-
ference on Soft Robotics (RoboSoft), May 15, 2023, https://doi
.org/10.1109/robosoft55895.2023.10122091.

94 STANFORD EMERGING TECHNOLOGY REVIEW

18. Stanford Zhao Lab, “Soft Intelligence Materials Laboratory,”
Stanford University, accessed August 30, 2023, https://zhaolab
.stanford.edu.
19. Stanford Autonomous Systems Laboratory, “Stanford ASL,”
Stanford University, accessed August 30, 2023, https://stanfordasl
.github.io.
20. Stanford Biomechatronics Laboratory, “Designing Robots to
Improve Mobility,” Stanford University, accessed August 30, 2023,
https://biomechatronics.stanford.edu.
21. Stanford Intelligent and Interactive Autonomous Systems
Group, “Iliad,” Stanford University, accessed August 30, 2023,
https://iliad.stanford.edu.
22. Matt Travers and Howie Choset, “Bioinspired Robots: Examples
and the State of the Art,” American Association for the Advance-
ment of Science, accessed August 30, 2023, https://www.science
.org/content/page/bioinspired-robots-examples-and-state-art.
23. Wyss Institute for Biologically Inspired Engineering, “A Self-
Organizing Thousand-Robot Swarm,” Harvard University,
August 14, 2014, https://wyss.harvard.edu/news/a-self-organizing
-thousand-robot-swarm.
24. Christoph Bartneck et al., Human-Robot Interaction: An Intro-
duction (Cambridge: Cambridge University Press, 2020), https://
www.human-robot-interaction.org.
25. Artificial Intelligence Lab, “Stanford Robotics Lab,” Stanford
University, accessed August 30, 2023, https://cs.stanford.edu
/group/manips/index.html; see also Oussama Khatib, “The Age of
Human-Robot Collaboration,” in ROMANSY-22—Robot Design,
Dynamics, and Control, ed. Vigen Arakelian and Philippe Wenger
(Rennes, France: Springer Cham, 2018).
26. Neil Savage, “Robots Rise to Meet the Challenge of Caring for
Old People,” Nature, January 19, 2022, https://www.nature.com
/articles/d41586-022-00072-z.
27. James Wright, “Inside Japan’s Long Experiment in Auto-
matic Elder Care,” MIT Technology Review, January 9, 2023,
https://www.technologyreview.com/2023/01/09/1065135/japan
-automating-eldercare-robots.
28. Gianmarco Veruggio and Fiorella Operto, “Roboethics: Social
and Ethical Implications of Robotics,” in Springer Handbook of
Robotics, ed. Bruno Siciliano and Oussama Khatib (Berlin: Springer-
Verlag, 2008), 1499–1524.
29. Cade Metz, “When the A.I. Professor Leaves, Students Suffer,
Study Says,” New York Times, September 6, 2019, https://www.
nytimes.com/2019/09/06/technology/when-the-ai-professor-leaves
-students-suffer-study-says.html.

07 Robotics 95