Autonomous agents can explore the environment around them when equipped with advanced hardware and software systems that help intelligent agents minimize collisions.These systems are developed under the term Artificia...Autonomous agents can explore the environment around them when equipped with advanced hardware and software systems that help intelligent agents minimize collisions.These systems are developed under the term Artificial Intelligence(AI)safety.AI safety is essential to provide reliable service to consumers in various fields such asmilitary,education,healthcare,and automotive.This paper presents the design of an AI safety algorithmfor safe autonomous navigation using Reinforcement Learning(RL).Machine Learning Agents Toolkit(ML-Agents)was used to train the agentwith a proximal policy optimizer algorithmwith an intrinsic curiositymodule(PPO+ICM).This training aims to improve AI safety and minimize or prevent any mistakes that can cause dangerous collisions by the intelligent agent.Four experiments have been executed to validate the results of our research.The designed algorithmwas tested in a virtual environment with four differentmodels.A comparison was presented in four cases to identify the best-performing model for improvingAI safety.The designed algorithmenabled the intelligent agent to perform the required task safely using RL.A goal collision ratio of 64%was achieved,and the collision incidents were minimized from 134 to 52 in the virtual environment within 30min.展开更多
Ensuring the safety of Artificial Intelligence(AI)is essential for providing dependable services,especially in various sectors such as the military,education,healthcare,and automotive industries.A highly effective met...Ensuring the safety of Artificial Intelligence(AI)is essential for providing dependable services,especially in various sectors such as the military,education,healthcare,and automotive industries.A highly effective method to boost the precision and performance of an AI agent involves multi-configuration training,followed by thorough evaluation in a specific setting to gauge performance outcomes.This research thoroughly investigates the design of three AI agents,each configured with a different number of hidden units.The first agent is equipped with 128 hidden units,the second with 256,and the third with 512,all utilizing the Proximal Policy Optimizer(PPO)algorithm.Importantly,all agents are trained in a uniformenvironment using the Unity simulation platform,employing theMachine Learning Agents Toolkit(ML-agents)in conjunction with the PPOalgorithm enhanced by an Intrinsic Curiosity Module(PPO+ICM).Themain aim of this study is to clearly highlight the benefits and limitations of increasing the number of hidden units.The results convincingly show that expanding the hidden units to 512 leads to a notable 50% enhancement in the agent’s Goal(G)and a substantial 50% decrease in the Collision(C)value.This study offers a detailed analysis of how the number of hidden units affects AI agent performance using the Proximal Policy Optimizer(PPO)algorithm,augmentedwith an Intrinsic Curiosity Module(ICM).By systematically comparing agents with 128,256,and 512 hidden units in a controlled Unity environment,the research provides valuable insights into the connection between network complexity and task performance.Theconsistent use of the ML-Agents Toolkit ensures a standardized training process,facilitating direct comparisons between the different configurations.展开更多
基金the United States Air Force Office of Scientific Research(AFOSR)contract FA9550-22-1-0268 awarded to KHA,https://www.afrl.af.mil/AFOSR/.The contract is entitled:“Investigating Improving Safety of Autonomous Exploring Intelligent Agents with Human-in-the-Loop Reinforcement Learning,”and in part by Jackson State University.
文摘Autonomous agents can explore the environment around them when equipped with advanced hardware and software systems that help intelligent agents minimize collisions.These systems are developed under the term Artificial Intelligence(AI)safety.AI safety is essential to provide reliable service to consumers in various fields such asmilitary,education,healthcare,and automotive.This paper presents the design of an AI safety algorithmfor safe autonomous navigation using Reinforcement Learning(RL).Machine Learning Agents Toolkit(ML-Agents)was used to train the agentwith a proximal policy optimizer algorithmwith an intrinsic curiositymodule(PPO+ICM).This training aims to improve AI safety and minimize or prevent any mistakes that can cause dangerous collisions by the intelligent agent.Four experiments have been executed to validate the results of our research.The designed algorithmwas tested in a virtual environment with four differentmodels.A comparison was presented in four cases to identify the best-performing model for improvingAI safety.The designed algorithmenabled the intelligent agent to perform the required task safely using RL.A goal collision ratio of 64%was achieved,and the collision incidents were minimized from 134 to 52 in the virtual environment within 30min.
基金supported by the United States Air Force Office of Scientific Research(AFOSR)contract FA9550-22-1-0268 awarded to KHA,https://www.afrl.af.mil/AFOSR/.
文摘Ensuring the safety of Artificial Intelligence(AI)is essential for providing dependable services,especially in various sectors such as the military,education,healthcare,and automotive industries.A highly effective method to boost the precision and performance of an AI agent involves multi-configuration training,followed by thorough evaluation in a specific setting to gauge performance outcomes.This research thoroughly investigates the design of three AI agents,each configured with a different number of hidden units.The first agent is equipped with 128 hidden units,the second with 256,and the third with 512,all utilizing the Proximal Policy Optimizer(PPO)algorithm.Importantly,all agents are trained in a uniformenvironment using the Unity simulation platform,employing theMachine Learning Agents Toolkit(ML-agents)in conjunction with the PPOalgorithm enhanced by an Intrinsic Curiosity Module(PPO+ICM).Themain aim of this study is to clearly highlight the benefits and limitations of increasing the number of hidden units.The results convincingly show that expanding the hidden units to 512 leads to a notable 50% enhancement in the agent’s Goal(G)and a substantial 50% decrease in the Collision(C)value.This study offers a detailed analysis of how the number of hidden units affects AI agent performance using the Proximal Policy Optimizer(PPO)algorithm,augmentedwith an Intrinsic Curiosity Module(ICM).By systematically comparing agents with 128,256,and 512 hidden units in a controlled Unity environment,the research provides valuable insights into the connection between network complexity and task performance.Theconsistent use of the ML-Agents Toolkit ensures a standardized training process,facilitating direct comparisons between the different configurations.
