Abstract:Abstract-Coordinated navigation by two cooperating sensor-equipped agents in a partially known static environment is investigated. Each agent observes a local part of the otherwise unknown environment and shares the gathered data with the other agents. In general, dynamic programming techniques suitably model the navigation problem, but are computationally hard to solve. We propose a combined dynamic and linear programming framework to circumvent the curse of dimensionality and establish in the process a firm … Show more
“…Let us denote by probe n the normalized difference between the front left and right probes as shown in Equation 16 and by probe abs = |probe n | the absolute value of this quantity.…”
Section: Control Unitsmentioning
confidence: 99%
“…Most of the research on autonomous pilots is directed toward piloting aircrafts [14,16,1,6], and cars [17]. Our approach targets motorcycles which have not been studied yet as extensively as the other types of vehicles and which represent a more challenging modeling problem.…”
In this paper we present an application of genetic algorithms to an autonomous pilot designed for motorized single-track vehicles (motorcycles). The pilot is implemented as a multi-agent application using a physical model of the motorcycle and is embedded in an interactive application. We compare the performance of the configuration obtained by genetic algorithms with the manual configuration of the pilot and with the performance of human players.The main contribution of the paper is proposing a model for piloting a single-track vehicle based only on perceptual information such as it would be observed by a human in the case of a real vehicle, and showing how the genetic algorithms can contribute efficiently to configuring the autonomous pilot.
“…Let us denote by probe n the normalized difference between the front left and right probes as shown in Equation 16 and by probe abs = |probe n | the absolute value of this quantity.…”
Section: Control Unitsmentioning
confidence: 99%
“…Most of the research on autonomous pilots is directed toward piloting aircrafts [14,16,1,6], and cars [17]. Our approach targets motorcycles which have not been studied yet as extensively as the other types of vehicles and which represent a more challenging modeling problem.…”
In this paper we present an application of genetic algorithms to an autonomous pilot designed for motorized single-track vehicles (motorcycles). The pilot is implemented as a multi-agent application using a physical model of the motorcycle and is embedded in an interactive application. We compare the performance of the configuration obtained by genetic algorithms with the manual configuration of the pilot and with the performance of human players.The main contribution of the paper is proposing a model for piloting a single-track vehicle based only on perceptual information such as it would be observed by a human in the case of a real vehicle, and showing how the genetic algorithms can contribute efficiently to configuring the autonomous pilot.
“…In past work [5], [6], we present an algorithm to compute optimal two-agent policies. In particular, the algorithm computes the optimal value function in two steps.…”
In this paper, we study spatially synchronous two-agent navigation on a structured partially unknown graph. The general edge cost statistics are given, and the agents gather and share exact information on the cost of local edges. The agents purpose is to traverse the graph as efficiently as possible. In previous work, we formulate the problem as a Dynamic Program, and exploit the structure of an equivalent Linear Program to compute the optimal value function. Here, we use the optimal policy to formulate a Markov chain with an infinite number of states whose properties we analyze. We present a method that computes the steady state probability distribution of the agent separation, exploiting the repetitive structure of the Markov chain as the agent separation goes to infinity. The results confirms and quantify the intuition that the less rewards, the more beneficial for the agents to spread out.
In this paper we introduce a visualization application for a vehicle simulation with an automatic pilot. The application is written in OpenGL and includes a model of a vehicle (a motorcycle) based on the physical laws of motion, including kinematics, gravitation, and friction. The vehicle can be controlled by the user through keyboard and mouse commands, as well as by an automatic pilot. The latter is implemented as a multi-agent probabilistic scheme using perceptual data.Our intention is to simulate the behavior of a human driver on the road. The performance of the automatic pilot is compared with that of a human player.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.