Is Nouvelle AI one software package?

No. It is an approach and historical research movement.

Does it reject all planning in every system?

The central historical critique concerns particular architectures; practical systems can combine reactive and deliberative elements.

Is a body always humanoid?

No. Embodiment can take many physical or simulated forms.

AI Glossary / AI and machine learning / Intermediate

Nouvelle AI

Definition

Nouvelle AI is a historically influential approach to artificial intelligence that emphasizes situated, embodied interaction and the development of useful behavior from relatively simple components. It is associated with Rodney Brooks's challenge to reliance on central symbolic world models.

What it means

Traditional symbolic AI often described a sequence of perceiving the world, building an internal representation, reasoning over it and then acting. Nouvelle AI questioned whether that architecture was the best starting point for creating adaptive behavior in real environments.

The approach emphasized building complete agents that could sense and act, beginning with simple capabilities and adding more. Intelligence was studied through the ongoing relationship between an agent and its environment rather than only through abstract problem solving.

Nouvelle means new in French, but the term refers to a historical movement rather than whatever AI happens to be newest today. It is not a single algorithm or a synonym for modern generative AI.

How it works

A typical behavior-based design links perception and action through interacting control layers. Lower-level behaviors can maintain basic functions such as avoiding obstacles, while additional layers support more complex activity. Subsumption architecture is a prominent related design.

The environment supplies information as the agent acts, reducing the need for one complete, centrally maintained world description in the kinds of tasks being studied. Designers test the integrated agent in realistic conditions and build capabilities incrementally.

This does not eliminate every form of internal state or representation. It challenges particular architectural commitments. More complex planning, memory and communication tasks may require additional mechanisms or hybrid approaches.

Three practical examples

A mobile-robot experiment builds obstacle avoidance before adding exploration behaviors.
A robotics course compares a reactive behavior-based controller with a central plan-based design.
An embodied-agent research project studies how simple sensor-action loops interact with the physical environment to produce useful behavior.

Think of it this way

Think of learning to move through a crowded room by continually responding to nearby people, rather than first building a perfect map of every person's future movement.

The analogy highlights responsive action, but people also use memory, goals and planning. Nouvelle AI's historical argument should not be stretched into a claim that all intelligent activity can be explained by immediate reactions alone.

Worked example

Combining simple behaviors

Imagine a toy mobile robot with three behaviors: stop when an obstacle is dangerously close, turn away from nearby obstacles and otherwise move forward to explore. The safety-related stop behavior takes priority when its condition is met.

In an open area, forward movement dominates. Near a wall, turning changes the sensor readings until the path clears. The robot can display useful navigation behavior without first constructing a detailed symbolic map of the whole room.

However, the same rules might cause repeated circling in a difficult layout. They do not guarantee reaching a distant named destination or remembering an earlier route.

This example illustrates behavior-based organization, not a complete implementation of Brooks's architecture. A real design must specify timing, arbitration, sensor uncertainty and failure handling, and test the complete system rather than assuming that individually sensible behaviors will combine successfully.

Illustrative case study

Fictional case: a robotics teaching prototype

Harbour Makers, an invented education team, builds a small robot for a controlled demonstration area. It begins with reliable stopping and obstacle avoidance, then adds simple exploration.

Students observe that the robot handles some local situations well but becomes trapped in a repeated path. They compare adding memory with redesigning the behavior interactions and discuss which capabilities require more than immediate sensor reactions.

The case illustrates the strengths and limits of an architectural idea. It does not claim autonomous operation in public spaces or a real commercial navigation result.

Common misconceptions and mistakes

Nouvelle AI is not the latest brand of AI, and it is not identical to deep learning. Its emphasis on embodiment does not require a human-shaped body. A behavior-based robot can still contain internal state and carefully engineered control logic.

The phrase “without representation” in historical arguments should be interpreted in its context. It does not establish that every task, including language and long-horizon planning, can be solved without any useful internal representation.

Business use

The business relevance is mainly architectural: useful autonomous behavior can depend on timely sensing, physical interaction and robust basic capabilities, not only on elaborate reasoning. This perspective can encourage teams to test integrated systems early.

For a practical project, identify which behaviors can be handled locally and which require memory, coordination or planning. Specify priority rules and test interactions under difficult conditions. Compare reactive and hybrid designs against the actual operating requirements. Historical inspiration should guide questions and experiments, rather than serve as a reason to reject all symbolic or model-based components.

Check your understanding

A reactive robot avoids obstacles but cannot reliably return to a previously visited place. Does local avoidance alone establish a complete navigation solution?

Answer: No. Returning may require memory, localization or planning beyond the demonstrated behavior. The system must be evaluated against the full task.

Related terms

Learn next

Learn next how subsumption architecture coordinates behavior layers. Then compare a local reactive controller with a hybrid controller that adds memory or planning, and identify which task requirements motivate the additional components.

Subsumption architecture

Sources

Primary source: Rodney Brooks, Intelligence without Representation, https://people.csail.mit.edu/brooks/papers/representation.pdf ; author publication record, https://people.csail.mit.edu/brooks/publications.html . Brooks's primary paper was consulted on incremental control layers, finite-state mechanisms, suppression and inhibition, and acknowledged limits of the approach. Robot example and teaching case are original.

Previous: Noun phrase Next: Novelty detection