Supervised Autonomous Vehicle Control Using Motion Planner Nudges
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current autonomous vehicle systems face challenges in smoothly transitioning control between autonomous and remote operator modes, and they often struggle to handle all situations optimally, necessitating a supervised autonomous mode where a supervisor can provide input to assist the vehicle without taking direct control.
Innovation Solution
The implementation of a supervised autonomous mode that allows a remote supervisor to provide inputs to the vehicle's motion planner through a nudge, which modifies the cost function based model, enabling the vehicle to adjust its trajectory and behavior in real-time while remaining in autonomous control.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a remote supervisor directly takes control of the vehicle, then the vehicle can handle complex situations optimally, but the transition between autonomous and remote control becomes problematic and disrupts operational stability
Solution Approach 1:
The patent introduces a supervisor input device as an intermediary between the remote supervisor and the vehicle's motion planner. Instead of directly taking control, the supervisor provides inputs that are converted into nudges modifying the cost function. This intermediary mechanism allows human guidance to be integrated smoothly into the autonomous system without disrupting the control transition, resolving the contradiction between handling complex situations and maintaining operational stability.
Solution Approach 2:
The autonomous vehicle system maintains self-service by keeping the motion planner in continuous control while incorporating supervisor inputs as modifications to the cost function. The system serves itself by autonomously processing both sensor data and supervisor nudges through its existing decision-making framework, eliminating the need for manual control takeover and ensuring smooth transitions while still benefiting from human oversight in complex situations.
2Adaptability or versatility
If the autonomous system operates independently without supervisor input, then operational stability is maintained, but the vehicle cannot optimally handle situations beyond its current capabilities
Solution Approach 1:
The patent merges the autonomous motion planner and human supervisor inputs into a unified control framework. The supervisor input device and the motion planner are combined such that supervisor nudges are integrated as modifications to the cost function rather than separate control commands. This merging enhances adaptability by incorporating human judgment while maintaining the existing autonomous system architecture, avoiding excessive complexity increase.
Solution Approach 2:
The cost function in the motion planner serves multiple functions: it processes sensor data for autonomous decision-making and simultaneously incorporates supervisor inputs as modifications. This multi-functionality allows the same computational framework to handle both independent autonomous operation and supervised guidance, enhancing versatility without proportionally increasing system complexity.
3Speed
If the supervisor provides continuous input to modify the cost function, then the vehicle can adapt to complex situations in real-time, but the computational load and system complexity increase
Solution Approach 1:
The supervisor input device provides partial action by offering selective nudges to modify specific elements of the cost function rather than continuously overriding the entire system. This partial modification approach enables real-time adaptability for critical decisions while keeping the overall computational load manageable, as only relevant cost function elements are adjusted based on supervisor input rather than recalculating the entire cost function continuously.
Data Source
AI summary
A method of operating an autonomous vehicle in a supervised autonomous mode is disclosed. A supervised autonomous mode can include the steps of receiving a vehicle data from a vehicle, presenting vehicle data at a remote station, receiving a supervisor input from a supervisor input device at the remote station, converting the supervisor input into a nudge for the motion planner of the vehicle associated with the received first set of vehicle data, wherein the nudge at least includes information on how to modify one or more elements of the cost function of the motion planner and transmitting the nudge from the remote station to the motion planner such that the nudge at least temporarily updates a cost function based model of the motion planner based on the nudge.