Visual Navigation Mapping With Multi-Scale Differentiable Planning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing navigation systems for agents in environments, such as robots, face inefficiencies and inaccuracies due to the lack of end-to-end differentiability in mapping and planning processes, which limits their ability to learn and improve environment understanding and action planning.
Innovation Solution
A navigation system that incorporates a differentiable mapper and planner, allowing for end-to-end training by backpropagating errors, and utilizing spatial scaling of environment maps to enhance action planning precision while maintaining accuracy, with the mapper generating belief and confidence data and the planner performing value iterations at multiple scales to select actions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional mapping and planning systems are used separately without end-to-end differentiability, then each subsystem can be developed independently, but the overall navigation accuracy and learning efficiency deteriorate due to inability to backpropagate errors across subsystems
Solution Approach 1:
The patent merges the mapping subsystem and planning subsystem into a unified end-to-end differentiable system. The mapping module generates environment maps that are directly fed into the planning module, which performs value iteration to select actions. This integration allows gradients to flow back through both modules during training, enabling joint optimization and improving navigation accuracy while maintaining manageable complexity through modular architecture design.
2Measurement precision
If high-resolution environment maps are used for planning, then action planning precision is improved, but computational efficiency deteriorates due to increased processing requirements
Solution Approach 1:
The patent segments the planning process into multiple stages with different spatial resolutions. The system performs value iteration at multiple scales, starting with coarser resolutions for global planning and progressively refining to finer resolutions for local action selection. This multi-scale approach maintains high planning precision for final actions while reducing overall computational burden by avoiding full high-resolution processing throughout the entire planning sequence.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Methods, systems, and apparatus, including computer programs encoded on computer storage media, for navigation using visual inputs. One of the systems includes a mapping subsystem configured to, at each time step of a plurality of time steps, generate a characterization of an environment from an image of the environment at the time step, wherein the characterization comprises an environment map identifying locations in the environment having a particular characteristic, and wherein generating the characterization comprises, for each time step: obtaining the image of the environment at the time step, processing the image to generate a first initial characterization for the time step, obtaining a final characterization for a previous time step, processing the characterization for the previous time step to generate a second initial characterization for the time step, and combining the first initial characterization and the second initial characterization to generate a final characterization for the time step.