Semantic Map Motion Planning for Mobile Robot Trajectory Generalization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for predicting future trajectories of mobile robots in dynamic environments, such as those with humans and other objects, face challenges in generalizing to new environments and handling heterogeneous datasets, often failing to converge to optimal reward function parameters due to differences in empirical feature counts across various semantic maps.
Innovation Solution
A computer-implemented method that uses semantic maps to determine reward function parameters by computing differences between expected and empirical mean feature counts, updating these parameters iteratively, and employing a backward-forward algorithm to simulate trajectories and infer occupancy maps, allowing adaptation to diverse environments without additional adaptation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If conventional inverse optimal control methods are used to determine reward function parameters, then the method can learn from demonstrated behaviors, but it fails to generalize well to new environments or heterogeneous datasets with different geometries
Solution Approach 1:
The patent transforms the reward function learning problem from directly learning parameters to learning occupancy priors that are environment-agnostic. By changing the parameter representation from environment-specific reward weights to universal occupancy probabilities, the system achieves both generalization across heterogeneous datasets and reliable convergence through iterative optimization of the occupancy prior distribution.
2Loss of information
If semantic maps are used for encoding reward function features, then future human motion can be predicted, but the approach does not generalize well to new environments with different geometries
Solution Approach 1:
The patent extracts the essential semantic information from environment-specific maps and reformulates it as environment-agnostic occupancy priors. By separating the geometric specifics from the semantic content, the system retains the predictive power of semantic maps while achieving generalization to new environments through the use of universal occupancy probability distributions.
Solution Approach 2:
The patent changes the parameter representation from semantic features tied to specific map geometries to occupancy priors that are geometry-independent. This transformation allows the same occupancy prior distribution to be applied across different environments, enabling generalization while preserving the semantic encoding capability for predicting human motion.
3Measurement precision
If reward function parameters are determined based on empirical feature counts from training data, then the model fits the training data, but it oscillates between local minima and fails to converge
Solution Approach 1:
The patent introduces a feedback mechanism where the occupancy prior is iteratively refined by comparing predicted trajectories with actual training trajectories. The system uses the difference between expected and empirical feature counts as feedback to update the occupancy prior, creating a closed-loop optimization process that converges to a stable solution by continuously reducing the discrepancy between predictions and observations.
Data Source
AI summary
A computer-implemented method for determining a motion trajectory for a mobile robot based on an occupancy prior indicating probabilities of presence of dynamic objects and/or individuals in a map of an environment. Occupancy priors are determined by a reward function defined by reward function parameters. The determining of the reward function parameters includes: providing semantic maps; providing training trajectories for each of semantic maps; computing a gradient as a difference between an expected mean feature count and an empirical mean feature count depending on each of the semantic maps and on each of the training trajectories, the empirical mean feature count is the average number of features accumulated over the provided training trajectories of the semantic maps, wherein the expected mean feature count is the average number of features accumulated by trajectories generated depending on the current reward function parameters; and updating the reward function parameters depending on the gradient.