A Deep Learning Based Behavioral Approach to Indoor Autonomous Navigation

Abstract

We present a semantically rich graph representation for indoor robotic navigation. Our graph representation encodes: semantic locations such as offices or corridors as nodes, and navigational behaviors such as enter office or cross a corridor as edges. In particular, our navigational behaviors operate directly from visual inputs to produce motor controls and are implemented with deep learning architectures. This enables the robot to avoid explicit computation of its precise location or the geometry of the environment, and enables navigation at a higher level of semantic abstraction. We evaluate the effectiveness of our representation by simulating navigation tasks in a large number of virtual environments. Our results show that using a simple sets of perceptual and navigational behaviors, the proposed approach can successfully guide the way of the robot as it completes navigational missions such as going to a specific office. Furthermore, our implementation shows to be effective to control the selection and switching of behaviors.

3D Scene Graph: A Structure for Unified Semantics, 3D Space, and Camera

“…One common application is that of Conditional Random Fields (CRF) [28] for semantic segmentation, often used to provide globally smooth and consistent results to local predictions [43,25]. In the case of robot navigation, employing semantic graphs to abstract the physical map allows the agent to learn by understanding the relationship between semantic nodes independent of the metric space, which results to easier generalization across spaces [42]. Graph structures are also commonly used in human-object interaction tasks [39] and other spatiotemporal problems [20], creating connections among nodes within and across consecutive video frames, hence extending structure to include, in addition to space, also time.…”

Deep Local Trajectory Replanning and Control for Robot Navigation

“…Similar to [8,9,10,12], we use deep learning [28] to parameterize a motion policy. This approach allows us to forgo hand-engineered features for sensor data.…”

“…In this work, we combine ideas from machine learning [8,9,10,11,12] and hierarchical planning [13] to improve reactive robot control. Our approach does not require handspecifying all the parameters of the reactive controller; instead, most parameters are optimized based on example navigation data through imitation learning [14,15,8].…”

A Behavioral Approach to Visual Navigation with Graph Localization Networks

“…In terms of our topological representation, our work is closely related to that of Sepulveda et al [40]. However, we do not rely on modifying the environment by introducing artificial landmarks, and we define a reduced set of primitive behaviors.…”

“…These modifications help to (1) facilitate the design of topological maps for realistic, human environments, and (2) increase the robustness of learned navigation behaviors given limited data. Furthermore, the work in [40] does not pose navigation as a graph traversal problem.…”