Personal tools
You are here: Home CG seminar 2020 Efficient Decision Making and Belief Space Planning using Sparse Approximations
« December 2019 »
December
SuMoTuWeThFrSa
1234567
891011121314
15161718192021
22232425262728
293031
Log in


Forgot your password?
 

Efficient Decision Making and Belief Space Planning using Sparse Approximations

Wednesday, December 4th, 2019, 16:10

Checkpoint 480

underline

Efficient Decision Making and Belief Space Planning using Sparse Approximations

Khen Elimelech, The Technion

Abstract:

These days, intelligent autonomous agents and robots can be found all around us. These agents share the same fundamental goal -- to autonomously plan and execute their actions. In order to achieve reliable and robust performance, these agents should account for real-world uncertainty. Also, problems, such as long-term autonomous navigation, active Simultaneous Localization and Mapping (SLAM), and sensor placement over large areas, often involve optimization of numerous variables. These settings require reasoning over high-dimensional probabilistic states, known as "beliefs". Appropriately, the corresponding problem is known as Belief Space Planning (BSP). The computational complexity of this problem can turn exceptionally high, thus making it challenging for online systems, or when having a limited processing power.
 
In the first part of this talk, we will formulate the BSP problem, and state-of-the-art solution methods. In the second part, we will introduce fundamental work towards efficient decision making via problem simplification. We will show that a wise simplification method can maintain the same action selection (“action consistency”), or one for which the maximal loss can be guaranteed. We will then practically apply these ideas to BSP problems, in which the problem can be simplified by considering a sparse approximation of the initial belief. Finally, we will demonstrate the benefits of the approach in the solution of a highly realistic active-SLAM problem, where we manage to significantly reduce computation time, with practically no loss in the quality of solution.
 
Document Actions