Local Grid Map Obstacle Avoidance for Dynamic Indoor Robots
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing mobile robot navigation systems face challenges in safely and efficiently navigating complex dynamic environments due to limitations in obstacle avoidance strategies, particularly with dynamic obstacles, where perception errors and sensor noise lead to uncertainty and inefficiency.
Innovation Solution
A dynamic obstacle avoidance method based on real-time local grid map construction, utilizing RGBD sensors, obstacle detection and semantic segmentation networks, and deep reinforcement learning frameworks to predict obstacle trajectories and send real-time speed instructions to the robot, improving navigation safety and efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If learning-based obstacle avoidance strategies using deep neural networks are used, then obstacle avoidance capability in environments with static obstacles is improved, but the method is greatly limited by quality of the label training dataset and distribution mismatch between training and test datasets
Solution Approach 1:
The system uses unsupervised learning to automatically learn obstacle representations and navigation policies from raw sensor data without requiring manually labeled training datasets. The robot autonomously explores the environment and learns obstacle characteristics through interaction, eliminating the need for external labeling services and reducing data quality constraints.
Solution Approach 2:
The patent transitions from supervised learning paradigms to unsupervised learning approaches, fundamentally changing the learning parameter regime. This involves shifting from labeled data distributions to unlabeled data distributions, and from fixed training-test splits to continuous online learning, thereby resolving distribution mismatch issues.
2Adaptability or versatility
If distributed multi-robot navigation methods are proposed for highly-dynamic unstructured environments, then navigation capability in dynamic environments is improved, but complex framework design and extensive online computations increase the perceived uncertainty of the whole system
Solution Approach 1:
The patent divides the navigation task into modular components: obstacle detection, trajectory prediction, and policy selection. Each module operates independently with well-defined interfaces, simplifying the overall framework while maintaining adaptability to dynamic environments. This modular architecture reduces system complexity compared to monolithic distributed multi-robot systems.
Solution Approach 2:
The system pre-computes multiple candidate trajectories and pre-identifies obstacle patterns before actual navigation decisions are required. This preliminary preparation reduces online computational burden during critical navigation moments, lowering perceived uncertainty while maintaining high adaptability.
3Loss of information
If it is assumed that nearby obstacles and environments have been completely observed, then uncertainty caused by perception errors is addressed, but sensor noise and environmental complexity still cause perception errors that cannot be safely handled
Solution Approach 1:
The patent incorporates safety margins and uncertainty buffers in the navigation policy. Instead of assuming complete observation, the system proactively plans for potential perception errors by maintaining larger safety distances and having backup avoidance strategies ready, thereby cushioning against the harmful effects of sensor noise and environmental complexity.
Solution Approach 2:
The system continuously monitors sensor inputs and navigation outcomes, using feedback to adjust its perception and avoidance strategies in real-time. This closed-loop approach allows the robot to adapt to actual environmental conditions and compensate for perception errors caused by sensor noise and complexity.
Data Source
AI summary
A dynamic obstacle avoidance method based on real-time local grid map construction includes: acquiring and inputting Red-Green-Blue-RGBD image data of a real indoor scene into a trained obstacle detection and semantic segmentation network to extract obstacles of different types and semantic segmentation results in the real indoor scene and generate 3D point cloud data with semantic information; according to the 3D point cloud data, extracting and inputting state information of a dynamic obstacle to a trained dynamic obstacle trajectory prediction model, and predicting a dynamic obstacle trajectory in the real indoor scene to build a local grid map; and based on a dynamic obstacle avoidance model, sending a speed instruction in real time to the mobile robot to avoid various obstacles during the navigation process.