notesum.ai

Published at October 23

Integrating Large Language Models for UAV Control in Simulated Environments: A Modular Interaction Approach

cs.CV

cs.AI

Released Date: October 23, 2024

Authors: Abhishek Phadke¹, Alihan Hadimlioglu¹, Tianxing Chu¹, Chandra N Sekharan¹

Aff.: ¹Texas A&M University - Corpus Christi

Arxiv: https://arxiv.org/abs/2410.17602v1

Refer to caption

Interaction Stream	Function
startMission	This starts the mission and calls the pre-defined interaction stream functions in the CEP-1.
getMissionCoordinates	This checks the start and end mission coordinates for the UAV agent.
senseEnvironment	This loads the pre-mapped 3D environment as an XYZ heightmap to sense environment dynamics.
getAgentPosition	This uses the superimposed local coordinate system on the XYZ environment heightmap shown in Figure 6 to generate the Agent position. 1s indicate presence of obstacle and 0s indicate no obstacle
moveAgent	This uses the agent model and control code in CEP-2 to adjust actuators and produce roll, pitch, and yaw motions.
avoidObstacle getObstacleDimensions executeAgentManeuver	This compares the obstacle dimensions from the XYZ heightmap and agent position to generate an avoidance path around obstacle dimensions. The executeAgentManeuver is a sub-call that asks for ½ or 3-second iteration control commands in the x, y, and z plane for the agent to continue motion in the suggested direction.