Grid Cell Navigation Codes for Direct-Route Action Selection

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing reinforcement learning systems for navigation tasks are inefficient in unfamiliar or changing environments, requiring excessive computational resources and failing to utilize direct routes.

Innovation Solution

An action selection system utilizing a grid cell neural network that processes velocity data to generate grid codes, integrating spatial location and heading direction, combined with an action selection network trained by reinforcement learning, to efficiently navigate environments.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If reinforcement learning systems use traditional action selection methods, then they can learn navigation policies, but training requires excessive computational resources and time

Engineering Contradiction:
Improvetraining efficiencyVSAvoidcomputational resources
Core Design Contradiction:
ProductivityVSUse of energy by moving object

Solution Approach 1:

The system segments the navigation problem into two independent components: (1) a geometry prediction network that learns spatial relationships and grid codes, and (2) an action selection network that uses these grid codes for navigation. This segmentation allows pre-computation of geometric features that can be reused across different navigation tasks, reducing redundant computation during training.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The geometry prediction network performs preliminary learning of spatial geometry and generates grid cell representations before the action selection network is trained. This preliminary action creates a structured spatial framework that accelerates subsequent navigation policy learning by providing pre-processed geometric features.

Inventive Principle:
Principle #10Preliminary action

2Manufacturing precision

If reinforcement learning systems explore environments thoroughly, then they can learn comprehensive navigation policies, but they fail to utilize direct routes and require excessive training time

Engineering Contradiction:
Improvenavigation accuracyVSAvoidtraining time
Core Design Contradiction:
Manufacturing precisionVSLoss of time

Solution Approach 1:

The system replaces traditional reinforcement learning exploration (mechanical trial-and-error) with a geometry-based computational approach. The grid cell representation provides direct spatial reasoning capabilities, allowing the network to compute optimal routes through geometric relationships rather than relying on extensive exploratory training.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

3Reliability

If reinforcement learning systems are trained in familiar environments, then they can achieve good navigation performance, but they fail to adapt to unfamiliar or changing environments

Engineering Contradiction:
Improvenavigation performanceVSAvoidenvironment adaptability
Core Design Contradiction:
ReliabilityVSAdaptability or versatility

Solution Approach 1:

The grid cell representation serves as a universal spatial framework that can represent multiple environments with different geometries and layouts. By learning geometric relationships in a environment-agnostic manner, the system can adapt to unfamiliar or changing environments without retraining, as the grid code structure remains valid across different spatial configurations.

Inventive Principle:
Principle #6Universality (Multi-functionality)

Data Source

PatentEP4446947B1Performing navigation tasks using grid codes
Publication Date: 2025.12.17 GDM HOLDING LLC
  • EP4446947B1 patent drawingFigure 1
  • EP4446947B1 patent drawingFigure 2
  • EP4446947B1 patent drawingFigure 3

AI summary

Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for selecting actions to be performed by an agent interacting with an environment. In one aspect, a system comprises a grid cell neural network and an action selection neural network. The grid cell network is configured to: receive an input comprising data characterizing a velocity of the agent; process the input to generate a grid cell representation; and process the grid cell representation to generate an estimate of a position of the agent in the environment; the action selection neural network is configured to: receive an input comprising a grid cell representation and an observation characterizing a state of the environment; and process the input to generate an action selection network output.