Semantic mapping and semantics-boosted navigation with path creation on a mobile robot

Towards S-NAMO: Socially-Aware Navigation Among Movable Obstacles

“…. ), guaranteeing negligible uncertainty as to the environment's state, or by on-board sensors only [13,19,31,33]. A limited number of propositions actually are able to deal with no prior or partial geometric knowledge [13,14,19,23,31,33], uncertainty as to object positioning [8, 13, 17-19, 26, 27, 29, 31, 33], object movability [13,[17][18][19]29] or object kinematics/physics [18,29] (Recapitulated in Table 1, columns 'Prior', 'Uncertainty' and 'Real-World').…”

“…World representation NAMO relies on an object-based representation of the world [2,5,6,8,[10][11][12][13][14][16][17][18][19]22,23,[26][27][28][29]31,33] (in opposition to an occupationspace-based one): in order to chose the best obstacle placement, it is necessary to reason about them as separate entities. Final placement selection is what actually tells NAMO apart from the well-known field of Rearrangement Planning [4].…”

“…In the literature, individual obstacles are simply assumed to have a boolean attribute of being movable or not. This attribute has until now been given as input [2, 5, 6, 8, 10-12, 17, 18, 26-28] (mainly for simulation-only algorithms), determined on-line from obstacle visual recognition results [19,22,29,31,33] or by manipulation tentative [13,14,16,23] (for real-world experiments). In order to be more realistic, other semantic information is used in more advanced approaches, like object kinematics and physics (mass, center of inertia, .…”

“…We must also say that rather few NAMO propositions have been applied in a real-world setting [8,13,19,22,29,31,33], and when they are, they always maintain a 3D representation of the world, though all NAMO algorithms execute their path finding subroutines in a 2D plane. 3D data is mainly used to allow for proper grasping of obstacles, but also for cross-plane rearrangement planning [33](e.g.…”

Semantic Path Planning for Indoor Navigation and Household Tasks

“…'fridge keeps milk fresh'. Using Semantics for path planning is still largely unexplored because only recently have SLAM algorithms with the help of neural networks been able to create a dense pixel by pixel encoded semantic map [5,11]. On their own semantic slam algorithms still don't output the required quality for semantic path planning, but with a visualisation tool, a human user could correct misclassifications where needed.…”