TL;DR:
When a robot fails, responsibility is shared across design (what it can do), operation (where and how it is used) and oversight (who watches and who can intervene).
A narrow sidewalk, a wheelchair and a robot that stops dead show how a sensible design choice can still exclude someone.
This is an ethics discussion, not legal advice. Rules differ by place and change over time.
A small delivery robot rolls along a sidewalk, meets a person in a wheelchair coming the other way, and stops. It was built to stop when something is in the way, and it did. But now the person cannot pass, and the robot sits there with its hazard light blinking. Nobody was hurt by a malfunction. Was anything wrong?
Why robot ethics is practical
Ethics here is not about whether robots will take over. It is the everyday question of how a machine that shares space with people should behave, who decides that, and who answers for it when it behaves badly. Engineers make these choices in the details: a stopping rule, a speed limit, a camera angle.
Worked example: a delivery robot blocking a sidewalk
Take a sidewalk 1.5 m wide. The robot is 0.6 m wide, drives along the middle, and a person in a wheelchair, 0.7 m wide, approaches (these sizes are for the example). Does the wheelchair user have room?
If the robot stays in the middle, the free width on each side is
(1.5 − 0.6) / 2 = 0.45 m. That is less than 0.7 m, so neither side fits the wheelchair.If the robot moves to one edge, the free width on the other side is
1.5 − 0.6 = 0.9 m. The wheelchair needs 0.7 m, leaving0.9 − 0.7 = 0.2 mto spare. It can pass.If the robot were 0.9 m wide (a bigger delivery model), the free width even at the edge would be
1.5 − 0.9 = 0.6 m, below 0.7 m. It would block the sidewalk completely, however it positions itself.So the outcome depends on a rule: pull to the edge when someone approaches, and a width small enough to leave a path.
Now ask the responsibility question for the failure where the robot stops in the middle. A review would look at three layers.
Design, operation and oversight
Design is what the robot was built to do. Did the designers consider wheelchairs, strollers, guide dogs and people who stop suddenly? Did they decide what the robot does when it cannot proceed: stop in place, pull to the side, or call for help? A stop-in-place rule is simple and safe against collisions, but it ignores the cost to others.
Operation is where and how the robot is used. A company that sends 0.9 m wide robots down 1.5 m sidewalks has made a choice that no code can fix. Routes, hours, speeds and the number of robots in one area are operating decisions.
Oversight is who watches and who can intervene. Is a remote person alerted when the robot has been stuck for 30 seconds? Can they move it? Can a passer-by push a button to make it yield? Do the people responsible keep records, learn from incidents and answer to the community or to a regulator?
Responsibility is usually shared. The robot is not the one that answers. The people and organisations that designed, deployed and supervised it do. Naming who is responsible before something goes wrong is part of doing the engineering well.
Accessibility
Public space is used by people with a wide range of abilities, and shared infrastructure should keep working for all of them. Robots that share sidewalks, shops and lifts should be tested with people who use wheelchairs, canes, guide dogs or strollers, and not only with engineers who walk briskly. Features such as sound or light signals should consider people who cannot see or hear them. Design rules and local regulations on accessible routes vary by place, so check the local ones.
Not legal advice: This article describes ethical questions in general terms. Who is legally liable for a robot's actions depends on the country, the product, the contracts and the facts. If you need an answer for a real situation, ask a qualified lawyer.
Where it goes wrong
Safe for the robot's goals, not for others. A rule that avoids collisions can still block people.
Nobody owns the failure. Each party assumes another is watching.
Hidden testing gaps. If the people who test never include someone with a mobility aid, the gap stays hidden.
Blaming the user. Saying the person should have moved ignores that the robot can be designed to yield.
Silent updates. Software changes that alter behaviour without review or notice.
No way to ask for help. A stuck robot with no visible contact or button leaves passers-by stuck too.
A practical checklist for a team: list how the robot can fail, who is affected in each case, what the robot does by default, who can override it, and what is recorded. Test the list with people who are affected, and update it after every incident.
How real robots use it
Teams use risk assessments, incident logs, remote monitoring and published safety cases to organise these questions. Standards bodies and regulators set requirements for particular kinds of robots, such as those described in collaborative robots and working safely near people. Wearable devices raise the same questions, as in exoskeletons, and so do robots that learn their behaviour from data, such as those that grasp objects they have never seen, where the designers cannot list every action in advance.
Try it yourself
Pick a machine you meet in daily life, such as an automatic door, a self-checkout or a lift. Write down who designed it, who operates it and who is responsible if it traps someone or fails to open. Then find one person who might struggle to use it and write how you would change the design. You have just done a small ethics review.
Frequently asked questions
Can a robot be responsible for what it does?
Today robots are tools. Responsibility rests with the people and organisations that build, deploy and supervise them.
Is stopping always the safest behaviour?
Stopping prevents some harms and causes others, such as blocking a path or an exit. A good design chooses a safe fallback for each situation.
Get started
