Robotics 101: the sense, estimate, perceive, plan, control loop
Post355 words19 likes
Describes a robot as a loop of sensing, state estimation, perception, layered planning and control atop hardware limits, and why ROS 2's node graph with QoS settings is a common systems framework.
Key claims
ISO 8373 frames a robot as a programmed, actuated mechanism with some autonomy
𝗥𝗼𝗯𝗼𝘁𝗶𝗰𝘀 𝟭𝟬𝟭
Robotics is about building machines that can sense, decide, and act in the physical world under uncertainty. A widely used standard framing (ISO 8373) describes a robot as a programmed, actuated mechanism with a degree of autonomy, moving within its environment to perform intended tasks.
Picture a robot as a loop that runs thousands of times per minute:
It starts with sensing. Cameras, LiDAR, radar, IMUs, encoders, each provides a partial, noisy view of the world and the robot’s own body. The robot then does state estimation, which is a fancy way of saying:
“given noisy measurements, where am I, how fast am I moving, and how uncertain am I about that?”
Without estimation, every downstream decision is built on sand.
Once the robot has a usable estimate, it does perception: detecting objects, free space, obstacles, and sometimes semantic labels (“this is a pallet,” “this is a human,” “this is a doorway”).
Perception is where modern ML helps a lot, but it’s also where robotics gets punished by edge cases: reflective surfaces, dust, rain, motion blur, odd geometry.
Then comes planning. Planning is typically layered:
➤ A high-level planner decides what to do next (go to aisle 7, pick item A, return to station)
➤ A motion planner decides how to move without collisions and within physical constraints
➤ A trajectory generator smooths the motion so it’s actually drivable (no impossible accelerations or jerks)
Control converts that plan into motor commands. This is the part many people underestimate: the robot doesn’t “move to a point,” it must continually correct itself while the world pushes back (wheel slip, payload shifts, friction changes, joints heat up).
Underneath all of it sits the hardware reality: mechanics, actuators, power, and thermal limits. A robot with great software but weak actuation or poor sensing becomes unreliable fast.
Modern robots are rarely a single monolith of software. They’re systems composed of modules that publish and subscribe to data streams.
This is one reason ROS 2 is widely used as a systems framework: it encourages a “graph of nodes” exchanging messages over topics, with tunable communication reliability via QoS settings.
1 image on X