Originally published in 2014; rebuilt in 2026. This post originally contained only a video. The video featured Hugh Herr’s work on advanced bionic limbs. The article has now been expanded into a public education resource about how designers and engineers turn human needs into working systems.
Quick Read
Designers and engineers change the world when they translate a real human need into a solution that works under physical, biological, economic and ethical constraints.
One-sentence answer: strong engineering begins with the receiver, identifies the real problem, builds a model, tests it against reality, learns from failure and keeps improving until the solution is useful, safe and reliable enough for real human use.
The original video: Hugh Herr and bionics
The video linked in the 2014 post is Hugh Herr’s TED talk on bionic prostheses. Herr, a biomechatronics researcher and double amputee, demonstrates how engineering can combine mechanics, electronics, sensing, control and human biomechanics to build artificial limbs that support walking, running, climbing and dancing.
The educational value is larger than the device itself. Bionics provides a clear example of an engineering problem in which the machine cannot be designed separately from the human being who will use it.
Design starts with a human need
An engineer can build something technically impressive that solves the wrong problem. That is why useful design begins by defining the receiver and the desired outcome.
- Who will use the system?
- What does the person need to do?
- What currently prevents that outcome?
- Which constraints cannot be ignored?
- What would success look like to the user?
For a prosthetic limb, success is not merely “the motor moved”. The person needs comfort, control, balance, durability, manageable weight, safe loading, reliable attachment and movement that works in everyday situations.
Engineering is constraint management
Real engineering rarely asks for one variable to be maximised without cost. A lighter structure may be weaker. A more powerful actuator may consume more energy. A more complex sensor system may become harder to maintain. Higher performance may increase cost.
The engineering task is therefore to find a workable region inside many constraints rather than chase one perfect number.
| Constraint | Question |
|---|---|
| Physics | Can the structure carry the required forces? |
| Energy | How long can the system operate? |
| Human factors | Is it comfortable and understandable to use? |
| Safety | What happens when something fails? |
| Cost | Can the intended user reasonably access it? |
| Maintenance | Can it be serviced over time? |
| Ethics | Does the design preserve dignity, agency and choice? |
The design loop
- Observe: understand the real environment and user.
- Define: state the problem precisely.
- Model: predict how a possible solution should behave.
- Prototype: build a testable version.
- Measure: compare actual performance with the prediction.
- Fail safely: identify what breaks without harming the user.
- Revise: change the design based on evidence.
- Validate: test again under realistic conditions.
The loop is powerful because it keeps the design answerable to the world. If reality disagrees with the model, the model has to change.
Bionics is a systems problem
A modern powered prosthesis can contain structural components, joints, sensors, processors, control algorithms, batteries and actuators. The user also contributes motion, intention, balance and adaptation.
No single component explains the final outcome. The useful behaviour emerges from the whole system working together.
This gives students a general engineering lesson: when a product appears simple at the interface, complexity may have been moved behind the interface through good design.
Sensing: the machine needs information
A robotic or bionic system cannot respond intelligently without information about its current state. Sensors may measure angle, force, acceleration, contact, pressure or other variables.
The basic control problem can be expressed simply:
- What state is the system in now?
- What state should it be in?
- What action should reduce the difference?
- Did the action actually work?
This receive-predict-act-check loop appears across engineering, from thermostats and drones to industrial robots and medical devices.
Control: movement must remain stable
A machine can have enough power and still be unusable if its control is unstable. In a prosthetic limb, movement must respond quickly enough to the user while avoiding oscillation, sudden unsafe motion or delayed correction.
Control engineering therefore turns raw hardware into coordinated behaviour. The goal is not maximum motion; it is appropriate motion at the right time.
The human-machine interface
Medical and assistive engineering requires an especially careful interface because the user is not outside the system. The device touches the body, changes movement and may become part of daily identity.
A strong design therefore asks more than “Can this device function?” It asks:
- Can the person understand and control it?
- Does the device fit safely?
- Can it be worn for meaningful periods?
- Does it reduce or create fatigue?
- Can the user trust what it will do next?
- Can the person override or stop it?
Failure is part of engineering
Students sometimes believe good engineers avoid failure. In reality, engineering depends on discovering failure early enough that it becomes information rather than harm.
A prototype should reveal weak assumptions. A stress test should identify where a component breaks. A simulator should expose control problems before a human depends on the system.
The aim is not to eliminate all failure from experimentation. It is to contain failure inside safe testing environments until the system is robust enough for wider use.
Safety has to be designed in
Safety cannot be added as a final label after the exciting work is complete. Engineers need to anticipate failure modes from the beginning.
- What if a sensor gives the wrong value?
- What if power is lost?
- What if the network connection disappears?
- What if the user moves unexpectedly?
- What if two subsystems disagree?
- Can the system fail into a safer state?
In medical and assistive technology, the cost of failure reaches a human body. That makes safety architecture a core part of the design.
Accessibility is part of usefulness
A device can work beautifully in a laboratory and still fail as a social solution if very few people can access it. Cost, clinical support, repair, training, geography and healthcare systems all affect whether technology reaches the person who needs it.
This is a useful distinction between invention and impact. Invention proves that something can exist. Impact requires the surrounding route to work too.
Repair versus enhancement
Bionic systems also raise an ethical boundary question. A technology may begin as a way to restore lost function and later develop capabilities that exceed ordinary human performance.
That does not make enhancement automatically good or bad. It does mean society needs to ask who receives the technology, who pays, what risks are acceptable and whether new capabilities create unfair or coercive expectations.
Engineering and AI
AI can improve perception, prediction, optimisation and control, but it does not remove engineering responsibility. A learning system still needs defined operating boundaries, reliable sensors, appropriate fallback behaviour and a clear human authority structure.
A model can recommend an action. The larger system must still determine whether that action is safe, authorised and suitable for the actual user.
What students can learn from bionics
| Subject | Learning route |
|---|---|
| Mathematics | Forces, geometry, optimisation, signals and modelling |
| Physics | Motion, torque, energy and stability |
| Biology | Muscles, joints, gait and nervous-system interaction |
| Computing | Sensing, control, embedded systems and algorithms |
| Design | User needs, prototyping and iteration |
| Ethics | Access, agency, enhancement and safety |
| English | Explaining complex systems clearly to non-specialists |
The deeper design principle
The best engineering is not impressive because it contains many parts. It is impressive because the parts disappear into a useful human outcome.
That is why Hugh Herr’s bionics work remains a strong educational example. It connects a human need with biomechanics, electronics, materials, computation, control and design—then returns the result to the human being who must live with it.
Source
The original video is Hugh Herr’s TED talk, The new bionics that let us run, climb and dance, published in 2014.
Historical note: this page began in 2014 as a single-video post. The 2026 revision keeps that original object and adds a durable engineering-learning layer around human needs, constraints, sensing, control, safety, iteration and responsible innovation.
A Deeper Reader: Eight Questions for Turning Bionics from an Impressive Device into Responsible Human Engineering
Bionics becomes educationally powerful when the machine stops being the centre of the story. A powered prosthesis can contain elegant mechanics, fast processors and sophisticated control, but none of those features is the final outcome. The final outcome is a person moving through ordinary life: standing, walking, climbing, working, resting, recovering balance, wearing the device for long periods and trusting what it will do next.
This is a useful discipline for every engineering student. Technical success is not the same as human success. A prototype can perform beautifully on a test rig and fail because it is uncomfortable, expensive, difficult to repair, unsafe on wet ground or too complicated for the intended user. The system has to return to the person.
1. What is the real engineering problem?
The weak version of the problem is “build a robotic leg”. The stronger version is “help a particular person achieve useful movement under real constraints”. The second definition immediately creates better questions. What movements matter? What loads occur? How does the person signal intention? What is comfortable? What happens after hours of use? How is the device charged, maintained and adjusted? Which failures could injure the user?
Defining the problem through the human task also prevents feature inflation. More sensors are not automatically better. More power is not automatically better. More software is not automatically better. Every added capability carries weight, cost, energy use, complexity and new failure modes.
The design question is therefore not “How advanced can we make this?” It is “What combination of capabilities gives this person the safest, most useful outcome?”
2. Which distinctions protect engineering from spectacle?
- Prototype is not product. A prototype proves a concept under selected conditions; a product must survive repeated real-world use.
- Movement is not mobility. A joint can move while the user still lacks balance, comfort, endurance or confidence.
- Performance is not accessibility. A highly capable device may reach very few people if cost, fitting, repair or clinical support is unavailable.
- Automation is not autonomy for the user. A system that makes more decisions can still reduce the person’s agency if override and control are poor.
- Restoration is not enhancement. Replacing lost function and exceeding typical human capability raise overlapping but different ethical questions.
- Laboratory success is not field reliability. Sweat, rain, uneven surfaces, battery ageing and daily wear change the operating environment.
- Accuracy is not safety. A model may be statistically accurate and still be unsafe if rare failures are severe.
These distinctions matter because engineering publicity naturally highlights best-case capability. Education should also show the surrounding conditions that make the capability dependable.
3. How does intention become movement?
A bionic limb has to bridge two worlds. The human produces intention through body movement, muscle activity, loading patterns or other signals. The device receives measurements, estimates state, selects a response and produces force. The person then senses the result and adapts. The control loop is therefore partly machine and partly human.
This creates a design challenge that ordinary machinery does not have in the same form: the user learns the machine while the machine may also adapt to the user. Good performance can emerge through repeated mutual calibration. That means fitting, rehabilitation and training may be as important as hardware specifications.
Delay matters. If a device reacts too slowly, movement can feel unnatural or unstable. Noise matters. If a signal fluctuates, the controller may misread intention. Uncertainty matters. A safe system should know when its estimate is weak and avoid pretending confidence is higher than it is.
The deeper lesson is that interface quality often determines whether sophisticated internal engineering becomes usable external capability.
4. What evidence would show that the design really helps?
A dramatic demonstration answers only a narrow question: can the system perform this movement under these conditions? Broader evidence asks whether the improvement survives across people, surfaces, time and daily use.
- Does walking require less effort?
- Are falls or near-falls reduced?
- Can the user navigate stairs, slopes and uneven surfaces?
- Does socket comfort remain acceptable over meaningful periods?
- How often does maintenance interrupt use?
- How does the system behave as the battery depletes?
- Does the user feel more independent?
- What harms or trade-offs appear after months rather than minutes?
Human outcomes can also conflict. A faster gait may increase fatigue. A lighter device may reduce structural margin. A more active controller may feel responsive to one user and intrusive to another. Engineering evaluation therefore needs several measures rather than one headline metric.
Where medical claims are involved, evidence quality matters further. Small demonstrations are useful for feasibility; larger clinical studies and longer follow-up answer different questions. Students should learn to match confidence to the evidence available.
5. Which counterexamples reveal hidden design failures?
The powerful device nobody wears. The actuator performs superbly, but the socket creates discomfort after an hour. Technical capability exists; human utility collapses at the interface.
The intelligent controller with a bad sensor. The algorithm is sophisticated, but the input becomes unreliable under sweat or unusual motion. The output fails because good reasoning cannot rescue bad state information.
The affordable prototype with expensive maintenance. Initial cost appears manageable, but replacement components and specialist servicing are unavailable locally. Accessibility fails across time.
The safe default that traps the user. A conservative failure mode prevents a dangerous motion but also leaves the person unable to move somewhere unsafe, such as a road crossing or staircase. Safety must consider the situation created by the fallback, not only the component protected.
Counterexamples are useful because they force the engineering team to widen the boundary of the system. What looked like a device problem becomes a user, environment, service and recovery problem.
6. What changes when AI enters the control loop?
Machine learning can help recognise gait phases, classify terrain, predict intention or tune control. That can make systems more adaptive. It also creates a new requirement: behaviour learned from data must remain inside acceptable physical and clinical boundaries.
A system should not receive permission to make a dangerous physical action merely because a model assigns it a high probability. Engineering safety can require hard limits independent of the learned model: torque ceilings, safe-stop conditions, confidence thresholds, watchdogs and human override.
Data quality also matters. If training data represent only a narrow population, performance may degrade for bodies, gait patterns or environments not well represented. A medical device cannot assume that “average user” is a complete human category.
The best AI-enhanced engineering therefore combines adaptive capability with explicit boundaries, transparent testing and continued human authority.
7. How should a student practise with the bionics case?
- Rewrite the problem. Replace “design a prosthetic leg” with a user-specific mobility objective and list the constraints that appear.
- Draw the control loop. Mark human intention, sensors, state estimation, controller, actuator, physical movement and human feedback.
- Run a failure-mode exercise. Choose one sensor, battery or mechanical component and ask what downstream behaviours fail when it degrades.
- Design two metrics. Pick one technical measure and one user-centred measure. Explain why neither is sufficient alone.
- Test the interface. Identify which information the user needs to understand or trust before relying on the device.
- Compare restoration and enhancement. Write separate ethical questions for each without assuming one answer fits both.
- Build an accessibility chain. Include manufacturing, fitting, training, maintenance, funding and long-term support.
- Write a bounded claim. Describe what one compelling demonstration proves—and what it does not prove.
This gives students a genuine engineering habit: every exciting capability should generate a second set of questions about reliability, access, evidence and failure.
8. What should return to the human being?
The final test is not whether the engineer can explain the mechanism elegantly. It is whether the person using the device gains something worth having: safer movement, more freedom, less effort, better access or greater confidence without unacceptable new burdens.
This is why responsible engineering is often quieter than invention culture suggests. The most valuable improvements may be better fit, easier charging, repairable components, clearer feedback, safer fallbacks or lower cost rather than another spectacular mode.
When technology touches a body, dignity belongs inside the specification. The person is not a platform on which engineers demonstrate capability. The technology exists to serve the person’s chosen life.
Next route: continue from bionics into biomechanics, prosthetics, rehabilitation engineering and human-centred design. Keep this page as the systems-level owner: human need → engineered capability → safe real-world use.