Socially Aware Robot MPC Using Predicted Trajectory Priors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Standard Information Theoretic Model Predictive Control (IT-MPC) techniques face challenges in crowded environments, where significant changes in optimal control sequences due to new goals or dynamic obstacles lead to poor convergence, making them inadequate for navigation of mobile robots.
Innovation Solution
An informed sampling process using a data-driven trajectory prediction model, such as a neural network, is employed to generate control trajectories aware of surrounding dynamic objects, combining IT-MPC with deep reinforcement learning to improve convergence by predicting control trajectories based on the current environment state and obstacle poses.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If standard IT-MPC uses local sampling around the previous control sequence, then the method performs well when optimal control changes slightly, but convergence becomes poor when optimal control significantly changes due to new goals or dynamic obstacles
Solution Approach 1:
The patent applies preliminary action by using a data-driven trajectory prediction model to predict future control trajectories before actual execution. This prediction informs the sampling process, allowing the system to prepare for upcoming changes in optimal control sequences due to new goals or dynamic obstacles, thereby improving convergence performance while maintaining adaptability to dynamic changes
Solution Approach 2:
The patent implements feedback by using the predicted control trajectories from the data-driven model to continuously update and inform the IT-MPC sampling process. The prediction model learns from past trajectories and environment states, providing feedback that adjusts the sampling distribution to align with the actual optimal control sequence, resolving the contradiction between reliable convergence and adaptive response
Data Source
Figure 1
Figure 2
Figure 3~4
AI summary
The invention relates to a computer-implemented method for determining a control trajectory for a robotic device (1), comprising the steps of: - performing (S1-S8) an information theoretic model predictive control applying a control trajectory sample prior (u*) in each time step to obtain a control trajectory for a given time horizon (tf); - determining (S21) the control trajectory sample prior (u*) depending on a data-driven trajectory prediction model which is trained to output a control trajectory sample as the control trajectory sample prior (u*) based on an actual state of the robotic device (1).