Trajectory Prediction With Evolving Intent and Adaptive Sampling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional trajectory prediction systems assume fixed intent over the prediction horizon, leading to performance limitations in modeling evolving discrete intent of moving objects, which affects accuracy and coverage.
Innovation Solution
A probabilistic hybrid discrete-continuous automaton (PHA) model learned as a deep neural network is used to infer high-level discrete modes and low-level samples, with adaptive sampling and sample selection techniques to predict the trajectory of moving objects, such as vehicles or pedestrians, improving accuracy and coverage.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If conventional trajectory prediction systems assume fixed intent over the prediction horizon, then the computational complexity is reduced, but the prediction accuracy and coverage deteriorate due to inability to model evolving discrete intent
Solution Approach 1:
The system transitions from assuming fixed intent to modeling evolving discrete intent dynamically. The PHA model sequentially infers high-level discrete modes (intent states) over the prediction horizon, allowing intent to change over time while maintaining computational tractability through structured probabilistic inference.
Solution Approach 2:
The patent replaces conventional trajectory prediction mechanisms with a deep neural network-based PHA model. The neural network learns the probabilistic transitions between discrete intent states and continuous trajectory dynamics, substituting traditional mechanical or rule-based prediction approaches with a learned probabilistic model that captures evolving intent.
2Device complexity
If conventional trajectory prediction systems assume fixed intent over the prediction horizon, then the system structure remains simple, but the coverage of evolving intent scenarios deteriorates
Solution Approach 1:
The PHA model introduces dynamic intent evolution by sequentially inferring discrete modes over time. Instead of assuming a single fixed intent state, the system models how intent can transition between different discrete states (e.g., from lane keeping to lane changing) while maintaining a structured approach that preserves system manageability.
Solution Approach 2:
The deep neural network-based PHA model serves multiple functions: it infers high-level discrete intent modes, generates continuous trajectory samples, and adapts to various moving object behaviors. This multi-functional approach improves coverage of evolving intent scenarios while consolidating complexity into a unified probabilistic framework.
3Measurement precision
If the discrete space grows exponentially to account for evolving discrete intent, then the prediction accuracy improves, but the computational burden increases significantly
Solution Approach 1:
The patent segments the exponentially growing discrete space into a structured sequence of high-level discrete modes inferred by the PHA model. Instead of considering all possible intent combinations simultaneously, the system breaks down the problem into sequential mode transitions, reducing computational burden while maintaining accuracy through probabilistic inference at each step.
Solution Approach 2:
The system changes the parameterization of the discrete space by using a probabilistic hybrid automaton model with learned transition probabilities. Instead of enumerating all discrete intent combinations, the PHA model represents the discrete space through a manageable number of intent states with probabilistic transitions, reducing computational complexity while preserving the ability to model evolving intent.
Data Source
AI summary
Systems and methods for predicting a trajectory of a moving object are disclosed herein. One embodiment downloads, to a robot, a probabilistic hybrid discrete-continuous automaton (PHA) model learned as a deep neural network; uses the deep neural network to infer a sequence of high-level discrete modes and a set of associated low-level samples, wherein the high-level discrete modes correspond to candidate maneuvers for the moving object and the low-level samples are candidate trajectories; uses the sequence of high-level discrete modes and the set of associated low-level samples, via a learned proposal distribution in the deep neural network, to adaptively sample the sequence of high-level discrete modes to produce a reduced set of low-level samples; applies a sample selection technique to the reduced set of low-level samples to select a predicted trajectory for the moving object; and controls operation of the robot based, at least in part, on the predicted trajectory.