Robot 101 · Chapter 01
What is a robot made of?
In one paragraph: Strip away the marketing language — "AI-native," "collaborative," "embodied intelligence" — and every robot, from a fixed industrial arm to a warehouse mobile robot to a walking humanoid, reduces to the same three hardware pillars: actuation (what moves), sensing (what perceives), and compute (what decides). Real-world robots sort into five families: industrial arms, cobots, mobile robots (AMRs/AGVs), legged and humanoid robots, and specialty or service robots built around one task. Whatever the family, the end-effector — the part that actually touches the world — is often the difference between a robot that works in a lab and one that works on a task. And on the bill of materials, actuation is typically the single largest line item.
The hardware trinity: what every robot reduces to
A robot is a mechanical system that converts energy into controlled physical motion, guided by sensors and computation. Whatever the marketing copy says, every robot reduces to three hardware pillars: actuation — the motors, gearboxes, and joints that move it; sensing — the cameras, force sensors, encoders, and IMUs that tell it (and its controller) what is happening; and compute — the onboard processors that turn sensor data into motor commands, typically running a control loop at hundreds to a thousand times a second. Everything else — the structural frame, the wiring harness, the enclosure, the battery — exists to serve these three.
The physical structure connecting the actuators is described as a kinematic chain: a series of rigid links joined at controlled points of motion. That chain's geometry defines how far the robot can reach, how fast it can move, and how precisely it can position its end-effector. The frame has to be stiff enough to resist the reaction forces of its own actuators without flexing, yet light enough to keep total mass and inertia manageable. Industrial arms use cast iron or welded steel for maximum stiffness; cobots and humanoids trend toward aluminium alloy and carbon fibre to cut weight. Frames are largely a structural commodity — the differentiating value sits in the active components: actuators, sensors, and the battery.
Industrial arms and cobots: the established base
Industrial robot arms are the original and still-dominant robot form factor in manufacturing: a fixed-base chain, six or seven degrees of freedom, bolted to a floor or gantry and fenced off in a safety cage because its motions are fast enough and its payload forces strong enough to injure anyone nearby. FANUC, Yaskawa Motoman, ABB, and KUKA are the long-standing volume leaders, with industry estimates putting combined annual shipments well into the hundreds of thousands of units. Strengths: extreme repeatability (around ±0.02 mm), high endpoint speed, and payloads from a few kilograms to well over a tonne, run 24/7. Weaknesses: zero mobility, no autonomy beyond a programmed trajectory, and the cost of the cage infrastructure around them.
Cobots (collaborative robots) are fixed-base arms designed to share workspace with people without a cage. The enabling technology is joint torque sensing (or current-based torque estimation) plus compliant control that lets the robot yield on contact, with payload and speed kept low enough to stay under recognised pain-and-injury thresholds for accidental contact. Universal Robots pioneered the category; FANUC's CRX, ABB's GoFa, Doosan's A-Series, and AUBO Robotics are among the main competitors today. Cobots are widely described as the fastest-growing segment of the established arm market, driven by small-batch, high-mix manufacturing where a full safety cage isn't economical.
Mobile robots: AMRs and AGVs
Mobile robots are wheeled or tracked platforms that move autonomously through indoor space. Autonomous Mobile Robots (AMRs) use SLAM-based navigation and dynamic obstacle avoidance to operate without a fixed physical guidepath or predefined track — they can replan around obstacles; Automatic Guided Vehicles (AGVs) follow a physical or magnetic track and are simpler and cheaper for a fixed route. Geek+, Hai Robotics, and Quicktron are prominent names in warehouse logistics; Pudu Robotics and Keenon are prominent in food-and-beverage and building delivery. Whichever the category, they share a defining engineering constraint.
That constraint is the battery. A delivery robot that can't make it back to its charging dock in time, or whose battery management system trips mid-corridor, fails its service commitment — whether it's carrying warehouse totes or a meal tray down a hospital corridor. That makes reliable, high-cycle battery packs with a communicative battery management system one of the most direct and recurring purchasing needs for anyone operating a fleet of mobile robots at scale.
Legged and humanoid robots: the frontier
Bipedal humanoids — two legs, a torso, two arms, and a head, as seen in platforms such as Tesla's Optimus, Figure's humanoids, Unitree's G1, and Fourier's GR series — are built to operate in spaces designed for people, without modifying the building: climbing stairs, opening doors, using tools and appliances made for human hands. The engineering complexity here is an order of magnitude above a wheeled robot: dynamic balance (the robot is inherently unstable and must compute corrective forces hundreds of times a second), whole-body coordination (moving an arm shifts the centre of mass, which changes what the legs must do to stay upright), and fall recovery (a humanoid that can't get up from a fall has little value operating unsupervised).
Quadrupeds — four-legged platforms such as Unitree's Go2, Boston Dynamics' Spot, and ANYmal — occupy the middle ground: greater low-speed/static stability than bipeds in many stances, and strong terrain capability, at the cost of giving up arm-like manipulation in most configurations. For many real-world service-robot applications — a hospital corridor, a warehouse floor, a care setting — the practical form factor today usually isn't a full bipedal humanoid, which is expensive, mechanically complex, and can feel unsettling in close quarters. It's more often a wheeled mobile base carrying an arm and a compact sensor suite, sized to the task at hand. The bipedal humanoid form earns its complexity when the task genuinely requires a human-shaped body moving through human-built spaces.
End-effectors: the last metre of the value chain
The end-effector is what an arm terminates in — the part that physically touches the world and therefore determines what tasks the robot can actually do. It is also the most task-specific component in the system, the one most often swapped when the application changes, and the one most likely to be damaged in use. Parallel-jaw grippers — two opposing jaws, driven by a screw, pneumatics, or a direct-drive motor — are the workhorse, succeeding on flat, accessible surfaces (boxes, bottles, cylindrical parts) and struggling with soft, irregular, or fragile objects. Vacuum (suction) grippers use negative pressure through one or more cups against a flat surface and dominate e-commerce and food-packaging logistics because they handle boxes and flat-packed items faster than jaw grippers, but they fail on porous, curved, wet, or oily surfaces.
Soft grippers — silicone, foam, or pneumatically inflatable fingers that conform to irregular shapes — trade force output (typically well under 20 N) for the ability to handle a ripe tomato, a raw egg, or a soft object without precise force control; they are a natural fit for a feeding or serving task in a service-robot context. Dexterous, multi-finger hands sit at the frontier: still expensive and mechanically complex relative to a well-tuned gripper, but improving quickly. Tool changers — quick-disconnect wrist mechanisms — let one arm swap between a gripper, a suction cup, and another tool in seconds, extending a single arm's task coverage without the cost of a dexterous hand.
Where the bill-of-materials money goes
Across almost every robot class, actuation is typically the single largest line in the bill of materials — a full humanoid alone can carry dozens of powered joints, each with its own motor, gearbox, encoder, and driver. Sensing (cameras, force/torque sensors, encoders, IMUs, and increasingly LiDAR) and compute (the onboard processing board and its power supply) are the next largest categories, followed by the battery and its management system for anything mobile, and the end-effector for anything that manipulates objects. These proportions shift a great deal by robot class — a fixed industrial arm spends almost nothing on mobility power, while an AMR fleet spends heavily on battery packs — so any specific percentage breakdown should be treated as an indicative industry figure, not a fixed rule, and checked against current supplier quotes before it drives a design decision.
Sourcing note. Asaptic sources validated robot components — joint modules, sensors, batteries, and end-effectors — from the Shenzhen ecosystem, delivered with test data, landed-cost math for your destination, and English documentation. Send a sourcing enquiry or see what we source.