Brain-Inspired AI Uses Cognitive Maps
·July 28, 2026
Summary: Researchers have developed a energy-efficient artificial intelligence model inspired by hippocampal brain mechanisms. The system mimics how the human brain constructs cognitive maps, performs stochastic computations, and utilizes compositional coding to solve complex planning tasks without exhaustive calculations.
Published as a proof-of-concept for alternative AI architectures, the model flexible reacts to novel environments without requiring retraining, consuming a fraction of the power required by conventional multi-layer neural networks or large language models. The technology offers a viable framework for local deployment in autonomous robotics, edge computing devices, and vehicles operating under strict energy constraints.
Key Facts
- Biological Energy Efficiency: The human brain operates complex planning and cognitive tasks on approximately 20 watts of power, contrasting sharply with the megawatt-scale power demands of modern large language models and deep neural networks.
- Hippocampal Architectural Principles: The model translates three specific neural mechanisms into algorithmic code: geometric cognitive maps, stochastic scenario generation, and compositional reusability of action components.
- Non-Exhaustive Search Strategy: Guided by cognitive maps, the algorithm evaluates randomly generated intermediate steps toward a goal rather than computing full solution paths, significantly reducing computational overhead.
- Zero-Shot Adaptability: The system adjusts to structural environmental changes or novel tasks dynamically without requiring retraining or parameter optimization.
- Validation Tasks: The brain-inspired architecture was successfully validated across three experimental benchmarks: 2D spatial navigation, abstract multidimensional orientation, and building-block silhouette assembly and disassembly.
Source: Graz University of Technology
The capabilities of large AI systems are constantly improving, but they consume a great deal of energy during training and operation. The human brain, by contrast, is extremely energy-efficient: it requires only around 20 watts.
Researchers at Graz University of Technology, in collaboration with international partners, have developed a novel, brain-inspired AI model that can plan flexibly and solve complex problems. In doing so, it consumes significantly less energy than multi-layer neural networks or large language models.
“The brain works in a completely different way to today’s AI systems,” says Wolfgang Maass from the Institute of Machine Learning and Neural Computation at Graz University of Technology. “We are trying to translate the way it works into algorithms and apply them to AI systems.”
Three mechanisms of the brain
Inspired by neuroscientific studies of the hippocampus, Wolfgang Maass and his colleague Yukun Yang have identified three mechanisms that the human brain employs in planning and problem-solving:
- the creation of cognitive maps, i.e. the transformation of relations between abstract objects into geometric relations between neural codes in the brain, that provides a “sense of direction” like a spatial map;
- stochastic neural computations, i.e. the ongoing generation of hypothetical scenarios and predictions;
- and compositional coding, i.e. the breakdown of information and action plans into reusable components.
The translation of these mechanisms into algorithms enables the Graz-based AI model – much like humans or animals – to conceive of and test possible approaches to solving complex problems without having to completely calculate them all the way to a solution.
If a randomly selected intermediate step points towards the intended goal – here, the cognitive map guides the system – this path is pursued. At its new position, the system again evaluates various options for action, thereby gradually getting closer to the goal.
“With this approach, our AI model can also react flexibly to changed or new situations without needing to be retrained,” says Yukun Yang.
Assembling and disassembling a silhouette
The researchers successfully tested the capabilities of their brain-inspired AI using three tasks: navigating a two-dimensional space, orienting itself in an abstract, multi-dimensional space, and assembling and disassembling a silhouette made up of various building blocks.
The researchers emphasise that their approach is not intended to rival today’s large language models, but provides the basis for an alternative approach for certain applications. “We are still at a relatively early stage of development,” says Wolfgang Maass. “But our work shows that powerful AI does not necessarily require huge data centres and enormous amounts of energy.”
Suitable for robots and edge devices
In the long term, such brain-inspired systems could be used in robots, autonomous vehicles or other edge devices – in other words, wherever AI needs to operate locally with limited energy supply.
Wolfgang Maass works as a key researcher at the Bilateral AI Cluster of Excellence. The current study was conducted in collaboration with Tsinghua University and the National Research Council of Italy.
Key Questions Answered:
Q: How do cognitive maps help the AI model save operational energy?
A: Standard AI systems often calculate every potential action sequence to completion to find an optimal solution, which requires vast memory and processing power. Cognitive maps transform abstract relationships into geometric coordinates, allowing the model to instantly assess whether a randomly generated step points toward the goal, discarding unpromising paths before spending energy calculating them fully.
Q: What is the role of stochastic neural computation in this architecture?
A: Stochastic computation introduces controlled randomness to generate hypothetical scenarios and predictions on the fly. Rather than running deterministic, brute-force searches, the system samples probabilistic paths, mimicking how biological neurons use noise and variability to rapidly sample potential solutions.
Q: Where are these brain-inspired AI models intended to be deployed?
A: While not designed to replace large language models in centralized data centers, this light-footprint architecture is optimized for edge devices, autonomous vehicles, and mobile robotics. These real-world applications require real-time adaptation, complex path planning, and low latency under strict battery and thermal limits.
Editorial Notes:
- This article was edited by a Neuroscience News editor.
- Journal paper reviewed in full.
- Additional context added by our staff.
About this AI and cognition research news
Author: Philipp Jarke
Source: Graz University of Technology
Contact: Philipp Jarke – Graz University of Technology
Image: The image is credited to Neuroscience News
Original Research: Open access.
“Neural sampling from cognitive maps enables goal-directed imagination and planning” by Hui Lin, Yukun Yang, Rong Zhao, Giovanni Pezzulo & Wolfgang Maass. Nature Machine Intelligence
DOI:10.1038/s42256-026-01254-4
canisgallicus 