Extending DTGolog to deal with POMDPs
Gavin Rens, Alexander Ferrein, Etienne van der Poel · 2008
For sophisticated robots, it may be best to accept and reason with noisy sensor data, instead of assuming complete observa-tion and then dealing with the effects of making the assump-tion. We shall model uncertainties with a formalism called the partially observable Markov decision process (POMDP). The planner developed in this paper will be implemented in Golog; a theoretically and practically ‘proven ’ agent programming lan-guage. There exists a working implementation of our POMDP-planner. 1.