What Makes a System a Robot?
A robot is a physical machine that can sense its environment, process that information, and take physical actions — all without constant step-by-step human instruction. The three elements are essential: sensing (gathering information), computation (deciding what to do), and actuation (physically doing it).
This distinguishes robots from automated machines. A conveyor belt performs the same action repeatedly with no sensing. A robot perceives its situation and adapts. A packaging robot that detects the size and shape of each incoming item and adjusts its grip accordingly is a robot. A fixed-speed conveyor is not.
Modern robots span an enormous range: robotic arms in automotive assembly lines that weld frames with sub-millimetre precision; agricultural drones that detect crop health from aerial imagery and apply variable-rate treatments; surgical robots like the da Vinci system that give surgeons enhanced precision for minimally invasive procedures; and warehouse robots like Amazon's Kiva fleet that navigate dynamic environments to fulfil orders.
The Sense-Plan-Act Loop
The foundational architecture of almost every robotic system is the **sense-plan-act loop**. It runs continuously:
**Sense**: Gather information about the robot's state and environment through sensors. Cameras capture visual scenes. LiDAR (Light Detection and Ranging) fires laser pulses and measures return time to build precise 3D maps. IMUs (Inertial Measurement Units) measure acceleration and rotation. Force/torque sensors detect contact forces at end-effectors. GPS provides global position. Each sensor has its own noise characteristics, latency, and failure modes — robust systems fuse multiple sensor modalities.
**Plan**: Given the current sensed state and the goal, compute what action to take. This is where most of the intelligence lives. For a simple path-planning problem, this might be a graph search algorithm like A* finding the shortest obstacle-free route. For a complex manipulation task, it might be a learned neural network policy that maps sensor observations to joint torques.
**Act**: Execute the planned action through actuators. Electric motors drive joints; pneumatic or hydraulic actuators provide high force; tendon-driven systems mimic biological muscles. The actuator translates computed intent into physical motion.
The loop runs at rates from 10Hz for slow manipulation tasks to 1000Hz for high-speed reactive control. The faster the loop, the better the robot can react to unexpected changes.
How AI Is Transforming Robotics
Traditional robotics used hand-programmed control systems. An industrial robot arm executing a weld follows a precisely pre-programmed trajectory. This works perfectly in structured environments that never change — but breaks the moment an unexpected object appears or a part is slightly misaligned.
Modern AI-driven robotics uses **learned policies** — neural networks trained to map sensor observations directly to actions. Reinforcement learning allows robots to discover effective behaviours through trial and error in simulation: a robot hand learning to spin a Rubik's Cube to any configuration took 100 years of simulated practice before being deployed on real hardware.
**Foundation models** are now being applied to robotics. Models like Google's RT-2 and Figure's collaboration with OpenAI train on internet-scale video data plus robot demonstration data, learning to interpret natural language commands and execute them on novel objects in novel environments. A robot trained this way can be told "put the apple in the bowl" and generalise to any apple, any bowl, without task-specific programming. This is the most significant paradigm shift in robotics since the introduction of programmable logic controllers.
The "sim-to-real gap" is the core challenge of learning-based robotics: policies learned in simulation often fail in the real world due to differences in physics, sensor noise, and visual appearance. Domain randomisation — training in thousands of randomised simulated environments — is the primary technique for bridging this gap.