Mobile Robot Motion Planning With Hypernetwork Constraint Functions
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing motion planning methods for mobile robots, such as vehicles, often rely on simple geometrical primitives or machine learning models that require significant computational effort and memory, leading to overconservativeness and inefficiencies in obstacle avoidance.
Innovation Solution
A computer-implemented method using a hypernetwork to generate parameter sets for a main network, which acts as a constraint function for motion planning, allowing for real-time obstacle avoidance with reduced computational complexity and overconservativeness.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If simple geometrical primitives or traditional machine learning models are used for motion planning, then obstacle avoidance can be achieved, but computational effort and memory requirements increase significantly
Solution Approach 1:
The patent replaces traditional computational approaches (geometrical primitives and conventional machine learning models) with a hypernetwork architecture that generates constraint functions more efficiently. This substitution reduces computational effort and memory requirements while maintaining obstacle avoidance capability through a different computational paradigm.
Solution Approach 2:
The patent changes the parameters and structure of the computational system by introducing a hypernetwork that dynamically generates constraint functions. This parameter change allows the system to achieve the same safety guarantees with reduced computational complexity by fundamentally altering how constraint functions are computed rather than using traditional methods.
2Reliability
If traditional machine learning models are used for real-time constraint generation, then obstacle avoidance is achieved, but overconservativeness increases and efficiency decreases
Solution Approach 1:
The patent substitutes traditional machine learning models with a hypernetwork-based system that generates constraint functions more efficiently. This substitution eliminates overconservativeness and improves navigation efficiency by using a fundamentally different computational approach that better balances safety and performance.
Solution Approach 2:
The patent introduces dynamic constraint generation through thehypernetwork, which adapts constraint functions in real-time based on the current state and environment. This dynamic approach replaces static or pre-computed constraints, allowing the system to maintain safety while improving efficiency by generating only the necessary constraints when needed.
3Measurement precision
If more computational resources are allocated to motion planning, then constraint accuracy improves, but computational costs increase
Solution Approach 1:
The patent replaces resource-intensive traditional computational methods with ahypernetwork architecture that achieves comparable or superior constraint accuracy with lower computational cost. This substitution maintains measurement precision while reducing energy consumption through a more efficient computational paradigm.
Solution Approach 2:
The patent uses thehypernetwork to generate constraint functions that effectively copy or approximate the behavior of more complex traditional models without requiring the full computational resources. This allows the system to achieve accurate constraints with reduced computational cost by using a streamlined representation.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
In order to improve motion planning in a mobile robot (10), such as a vehicle, the invention proposes a method comprising a sensor system (22) capturing sensor data, wherein the sensor data are indicative of an environment (12), that contains an obstacle (14); and a control module processing the sensor data to obtain at least one binary cost map (54), wherein each binary cost map (54) is indicative of occupied and unoccupied portions of the environment (12); processing each binary cost map (54) with a hypernetwork (32) that generates a parameter set (θ) for a main network (34) based on the binary cost map (54). The parameter set (θ) causes the main network (34) to generate constraint data from a motion related state vector (x) of the mobile robot (10). After the main network (34) is set up with the parameter set (θ) the control module generates a control signal (46) based on a command signal (44), a current state vector (x) and the constraint data, and the control signal (46) is generated to cause the mobile robot (10) to move through the environment (12) along a robot trajectory with the proviso that each obstacle (14) is avoided.