Scott Davies, Andrew Ng, and Andrew Moore. Applying online search techniques to continuous-state reinforcement learning. In Proc. of the Fifteenth National Conference on Artificial Intelligence (AAAI-98), 1998.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....for generating a path that obeyed more of the kinematic constraints of the mobile robot, replanning as necessary after every control. However, this approach still contains approximations of the mobile robot as a point, and also does not respect the dynamic constrains of the robot. Davies et al. [5] propose an algorithm that is very similar in spirit to our mobile robot application; however, while they project a low dimensional approximate value function solution to a higher space, they do not explicitly search for policies that minimize expected reward. Instead, they use a heuristic search ....

Scott Davies, Andrew Ng, and Andrew Moore. Applying online search techniques to continuous-state reinforcement learning. In Proc. of the Fifteenth National Conference on Artificial Intelligence (AAAI-98), 1998.

....for generating a path that obeyed more of the kinematic constraints of the mobile robot, replanning as necessary after every control. However, this approach still contains approximations of the mobile robot as a point, and also does not respect the dynamic constrains of the robot. Davies et al. [5] propose an algorithm that is very similar in spirit to our mobile robot application; however, while they project a low dimensional approximate value function solution to a higher space, they do not explicitly search for policies that minimize expected reward. Instead, they use a heuristic search ....

Scott Davies, Andrew Ng, and Andrew Moore. Applying online search techniques to continuous-state reinforcement learning. In Proc. of the Fifteenth National Conference on Artificial Intelligence (AAAI-98), 1998.

.... the boundaries become less distinct between artificial intelligence and fields more reliant on continuous mathematics, such as control engineering, it is being recognized that heuristic search methods can play useful roles when applied to problems with infinite state spaces (e.g. Boone [1997] Davies et al. 1998]) However, the theoretical properties of heuristic search algorithms differ greatly depending on whether the state space is finite or infinite. For finite state space problems, a variety of wellunderstood algorithms are available to suit different needs. For example, the A algorithm finds ....

S. Davies, A. Ng, and A. Moore. Applying online search techniques to continuous-state reinforcement learning. In Proceedings of the Fifteenth National Conference on Artificial Intelligence (AAAI-98), pages 753--760, 1998.

....results require a parameterized stochastic policy. In comparison, we are able to use domain knowledge to perform gradient descent on a deterministic policy, because we will show that in section 4 that the reward function for our domain contains the necessary gradient information. Davies et al. [6] proposed an algorithm that is very similar in spirit to our mobile robot application; however, while they projected a low dimensional approximate value function solution to a higher space, they did not explicitly perform policy search to minimize expected reward. Instead, they use a heuristic ....

Scott Davies, Andrew Ng, and Andrew Moore. Applying online search techniques to continuous-state reinforcement learning. In Proceedings of the Fifteenth National Conference on Artificial Intelligence (AAAI-98), 1998.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC