What is a Robot?
A robot is a machine that can sense its environment, process that information and take physical action in the world. That three-step loop (sense, plan, act) is the defining characteristic of every robot ever built, from the earliest factory arms of the 1960s to the humanoid machines walking out of research labs today.
This lesson establishes what robots are, how they are classified, where they came from and why robotics presents challenges that pure software AI does not.
The Sense-Plan-Act Cycle
Every robot, regardless of how simple or sophisticated, operates around the same fundamental loop:
Sense: The robot gathers information about its environment and its own state using sensors: cameras, microphones, LIDAR, accelerometers, force sensors and more. Without sensing, a robot is blind to the world it is supposed to act in.
Plan: The robot processes sensor data to decide what to do next. This is where intelligence lives. In early robots, planning meant executing a pre-programmed sequence of instructions. In modern robots, planning involves computer vision, probabilistic reasoning, learned policies and real-time optimisation.
Act: The robot executes its decision using actuators: motors, hydraulic cylinders or pneumatic systems that produce physical movement. Action is where the robot leaves an impression on the physical world.
The loop runs continuously. The robot senses, updates its internal model of the world, plans the next action, executes it and immediately begins sensing again. For a drone maintaining position in a gust of wind, this loop may run hundreds of times per second. For a surgical robot making a precise incision, each iteration requires careful deliberation. The timescale varies enormously, but the structure is universal.
Types of Robots
Industrial robots are the most widely deployed robots in the world. These are fixed robotic arms found on manufacturing lines, performing tasks like welding car bodies, spray-painting panels, assembling electronics and moving heavy components. They are fast, precise, tireless and designed for highly structured, predictable environments. A traditional industrial robot does exactly what it is programmed to do and fails the moment anything departs from expectation.
Collaborative robots (cobots) are a newer category designed to work alongside human workers rather than in caged-off areas. They are slower and carry lighter loads than industrial robots, but they have sensors that detect proximity to humans and stop before causing injury. Companies like Universal Robots and FANUC have made cobots increasingly accessible to small and medium-sized businesses.
Mobile robots navigate through environments rather than being bolted to a fixed position. This includes warehouse robots like those used in Amazon fulfilment centres, hospital delivery robots that ferry medications and supplies between floors and agricultural robots that move through fields. Navigation is much harder than fixed-arm manipulation because the robot must continuously localise itself and avoid obstacles.
Aerial robots (drones and UAVs) operate in three-dimensional airspace. Their widespread commercial availability has made them the most visible category of robots to the general public. Uses range from photography and surveying to package delivery, agricultural spraying and military applications.
Humanoid robots are machines built in the shape of a human body: bipedal, with arms and hands. The humanoid form is appealing because the world is designed for humans (tools, doors, stairs, vehicles), so a robot with a human form can, in principle, operate in any environment humans occupy. Boston Dynamics' Atlas has demonstrated remarkable agility. Tesla's Optimus and Figure's Figure 01 are targeting factory and warehouse labour.
A Brief History
The history of robotics is longer than most people realise.
1961: The first industrial robot, Unimate, was installed on a General Motors assembly line in New Jersey. It weighed two tonnes and performed die casting and welding. This was entirely pre-programmed: no sensing, no adaptation.
1970s: Stanford's Shakey was the first robot to reason about its own actions. It could perceive its environment through a camera, plan a sequence of steps to achieve a goal and execute the plan. Slow by modern standards, but a conceptual leap.
1980sā1990s: Industrial robots proliferated across automotive and electronics manufacturing. Japan led the world in deployment. Robots became faster, more precise and cheaper.
2002: iRobot's Roomba brought robots into homes as consumer products. Not sophisticated, but enormously important for normalising human-robot coexistence.
2004ā2005: DARPA Grand Challenge. Self-driving vehicles competed to navigate desert terrain autonomously. In 2004, no vehicle finished. In 2005, five vehicles completed the course. This catalysed serious investment in autonomous driving.
2011: IBM's Watson won Jeopardy. Not a robot, but it demonstrated that AI systems could operate in unstructured, natural-language domains: a capability that would eventually be crucial for robot intelligence.
2016 onwards: Boston Dynamics released videos of Atlas and Spot that spread virally, demonstrating locomotion capabilities that seemed impossible. Deep learning started being applied to robot perception, grasping and control.
Present: Humanoid robots are leaving the lab. Foundation models trained on internet-scale data are being adapted to generate robot actions. The field is moving faster than at any point in its history.
What Makes Robotics Different from Pure Software AI
Building a chess-playing AI or a language model is extraordinarily difficult. But it is a problem that exists entirely in a controlled, digital world. The rules are fixed. Every input is clean and well-defined. There is no friction, no unexpected dust on the sensor, no object that shifted two centimetres since the last frame.
Robotics adds an entirely different category of challenge: the physical world.
The world is unstructured. Objects are not perfectly positioned. Lighting changes. Floors are uneven. Things fall over. A package that is slightly crumpled does not match the training image. Software AI never has to deal with any of this.
Sensing is noisy. Every sensor introduces uncertainty. A camera delivers blurry frames. A LIDAR point cloud has missing data where glass windows reflected the beam. An IMU drifts over time. The robot must make good decisions despite imperfect information.
Actions have physical consequences. A wrong prediction in a language model produces an unhelpful sentence. A wrong action by a robot may damage the object being handled, the robot itself or, most seriously, a person nearby. The stakes of being wrong are fundamentally different.
Real time matters. A robot interacting with a moving human must respond in milliseconds. There is no pause button while the model runs inference.
Deployment diversity is extreme. A software model runs on a server. A robot might need to operate in a desert at 50Ā°C, a freezer at ā25Ā°C, an underwater hull inspection or a surgery suite with sterile requirements. Hardware must survive the environment.
These challenges are why robotics has historically lagged behind AI in software domains: and why progress is now accelerating as better sensors, faster compute and more capable AI come together.
Quiz: Describe the sense-plan-act cycle in your own words. Name two ways that robotics is fundamentally more difficult than building AI systems that operate purely in software.