Predicted Occupancy Grid Map Generation via Recurrent Neural Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for generating occupancy grid maps (OGMs) in dynamic and unstructured environments are inadequate for providing accurate and reliable path planning for autonomous devices, as they fail to effectively account for moving objects and require frequent trajectory adjustments, leading to safety and efficiency issues.
Innovation Solution
A machine learning-based system using a neural network with encoders, recurrent neural networks (RNNs), and classifiers generates predicted OGMs by extracting features from input OGMs and motion-flow information to predict changes in the environment, allowing for efficient and safe path planning without the need for labeled datasets.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If a single OGM is used to represent the environment at a given time point, then path planning can be performed efficiently, but the planning is only valid if the environment is assumed to be static which does not hold for dynamic environments
Solution Approach 1:
The system performs preliminary action by predicting future OGMs at multiple time steps ahead of actual navigation needs. The recurrent neural network generates predicted OGMs for future time steps based on current sensor data, allowing the path planning system to anticipate environmental changes before they occur. This enables the autonomous device to plan paths that remain valid throughout the predicted time horizon without requiring frequent replanning.
Solution Approach 2:
The system applies dynamics by using a recurrent neural network that continuously updates predictions based on new sensor observations. The OGM prediction system dynamically adapts to changing environments by processing sequential sensor data and generating updated future predictions, allowing the path planning to remain reliable even as the environment evolves over time.
2Measurement precision
If the OGM is updated frequently for each sensor data reading, then the environment representation remains current, but the planned trajectory requires frequent modifications which may be unsafe and require many unnecessary steering corrections
Solution Approach 1:
Instead of reacting to each new sensor reading by immediately updating the path, the system performs preliminary prediction of environmental changes over multiple time steps. The RNN forecasts future OGMs, allowing the path planning system to evaluate trajectory validity over an extended time horizon and make stable, infrequent adjustments only when necessary.
Solution Approach 2:
The system implements feedback by using the recurrent neural network to continuously monitor environmental changes and predict their impact on the planned trajectory. The predicted OGMs provide feedback about future environmental states, enabling the system to maintain trajectory stability by only modifying the path when predictions indicate actual conflicts with obstacles or changes in the environment.
3Reliability
If object representations in the OGM are enlarged to account for movement, then moving objects are better accounted for, but free spaces that can be used are blocked by the enlarged object representations
Solution Approach 1:
The system replaces static enlarged object representations with dynamic predictions. The recurrent neural network predicts the actual future positions of moving objects at multiple time steps, allowing the path planning system to account for object movement accurately without permanently blocking free spaces. The predicted OGMs show where objects will be in the future, not where they are currently, preserving available navigation space while maintaining reliability in accounting for moving objects.
Data Source
AI summary
Methods and systems for generating a predicted occupancy grid map (OGM) over at least one future time step are described. The system include a first encoder for extracting OGM features from an input OGM in a current time step. The system also includes a recurrent neural network for generating a corrective term from at least the OGM features, wherein the corrective term represents predicted change to the input OGM, and wherein the corrective term is applied to the input OGM to generate a corrected OGM. The corrected OGM represents features corresponding to occupancy of the environment in a first future time step. The system also includes a classifier for converting the corrected OGM to the predicted OGM for the first future time step. The predicted OGM is fed back as input for performing generating a predicted OGM for a second future time step.