Is situated robotics the same as behavior-based robotics?

No. Behavior-based control is an architectural approach; situated robotics is a broader perspective on interaction.

Can a situated robot use a map?

Yes. Situated analysis does not by itself prohibit maps or planning.

Why does the environment matter if the code is unchanged?

It changes observations, available actions, and the consequences of those actions.

AI Glossary / Reinforcement learning and robotics / Intermediate

Situated robotics

Definition

Situated robotics studies and designs robots as agents whose behavior arises through ongoing interaction with their environment. It emphasizes the relationship among sensing, action, the robot's body, and the surrounding world. Historically, it is associated with approaches that questioned reliance on complete internal world descriptions before acting. Situatedness does not name one algorithm, and it should not be reduced to simply placing a robot at a physical location.

What it means

A robot changes the world by acting, and those changes alter what it senses next. Its capabilities also depend on its body and surroundings: wheel traction, sensor placement, corridor geometry, and available objects can shape behavior as much as the control software.
The situated perspective treats these relationships as central to understanding performance. A controller tested only on idealized inputs may miss difficulties that appear when sensing and movement interact. The same program can behave differently in different environments because its inputs and action consequences are different.

How it works

Begin with the complete task in its intended environment. Identify what the robot can sense, how it can act, and how the environment changes in response. Make timing and physical constraints part of the design rather than assuming they will be solved elsewhere.
Build and evaluate the sensing - action loop. A robot repeatedly observes, acts, and observes again, adjusting behavior from current interaction. Reactive and behavior-based architectures are common historical examples, but situated analysis can also inform systems with memory, maps, planning, or learning.
Examine how the environment supports or disrupts behavior. A wall can provide a useful reference for navigation, while reflective surfaces can confuse a sensor. Features that simplify a task in one location may be absent in another.
Test the integrated system across meaningful changes in conditions. Observe failures caused by delays, uncertainty, physical properties, and interactions among components. Simulation can support this work, but assumptions about sensing and dynamics need scrutiny before conclusions are transferred to a physical robot.

Three practical examples

A wall-following robot uses current distance measurements to adjust its movement as corridor geometry changes.
A manipulation system exploits contact with a fixture to align an object, reducing the need to infer its pose from vision alone.
A mobile-robot team studies how sensor placement and body dimensions affect the controller's behavior in cluttered spaces.

Think of it this way

Think of cycling along a winding path. You do not simply execute a fixed sequence of handlebar movements: the path, bicycle, balance, and visual feedback continually affect one another. Situated robotics brings that whole interaction into the explanation of behavior, rather than examining the decision rule in isolation.

Worked example

Consider an invented simplified distance-control example. A simulated robot should remain 1 meter from a wall. Its current distance is d, and each update changes that distance by u = 0.5 × (1 − d). Assume the next distance is exactly d + u, with no delay or disturbance.
Starting at d = 1.8 meters, the first command is 0.5 × (1 − 1.8) = −0.4 meters, so the next distance is 1.4. The next command is −0.2, giving 1.2. Then −0.1 gives 1.1. The error halves on each update.
Now suppose the physical response is only half the requested change because the simplified actuation model was wrong. The update becomes d_next = d + 0.25 × (1 − d). Starting again at 1.8, the distances become 1.6, 1.45, and 1.3375. The error now shrinks by a factor of 0.75 rather than 0.5 each step.
The software rule is unchanged, but the behavior differs because the action - environment relationship differs. This toy scalar calculation illustrates coupling and model assumptions, not a complete wall-following controller. It omits heading, motion geometry, obstacles, sensing error, and operating limits.

Illustrative case study

In this fictional case, Hazel Robotics demonstrates a corridor-following robot in a clean test area. The same controller behaves inconsistently in a second corridor with different wall surfaces and floor conditions.
The team investigates the entire loop rather than changing only the high-level decision rule. It finds that sensor readings and motion response differ from the original test assumptions. It adjusts the implementation and evaluates whether the changes improve behavior across both settings.
The project report describes the environments and physical conditions alongside the software version. It treats dependable interaction as an empirical property of the complete system, not as something established by the controller's name.

Common misconceptions and mistakes

Situated robotics does not mean that all internal representations are forbidden. Strong historical arguments against particular representational approaches should not be converted into a universal rule for every situated robot.
Another mistake is equating situatedness with having a body alone. Embodiment concerns the body's role; situatedness emphasizes the agent's ongoing relation with its environment. They are closely connected but highlight different aspects.
A successful demonstration in one setting also does not guarantee transfer. The robot may depend on environmental regularities that were never explicitly documented. Identifying those dependencies is part of responsible evaluation.

Business use

The perspective helps organizations understand why a robot that works in a laboratory may struggle at a customer site. It encourages evaluation of the integrated application, including physical layout, sensing conditions, timing, and human activity.
Document operating assumptions and test changes to them. Compare actual task outcomes across representative environments, and distinguish a software defect from a mismatch between the system's capabilities and the conditions in which it is used.

Check your understanding

How quickly does the ideal example's error shrink?

Answer: It halves each update.

What changes under the weaker actuation response?

Answer: The error is multiplied by 0.75 each update.

Why is the toy calculation not a full navigation controller?

Answer: It omits the robot's movement geometry and many sensing and operating constraints.

Related terms

Behavior-based robotics
Reactive control
Subsumption architecture
embodied intelligence
perception–action loop
affordance
robot–environment interaction

Learn next

Learn next how embodiment and environmental structure influence control. Compare reactive and hybrid architectures, then examine how integrated experiments reveal dependencies that isolated component tests can miss.

Sources

Brooks, Intelligence Without Representation, original paper hosted by MIT: https://courses.csail.mit.edu/6.803/pdf/brooks.pdf Brooks, A Robust Layered Control System for a Mobile Robot, MIT report record: https://dspace.mit.edu/entities/publication/362a32d0-a821-4a9d-a5ff-9e2b71c5539d

Previous: Situated approach (artificial intelligence) Next: Situation calculus