Ðåñßëçøç xiii List of Tables vi List of Figures viii List of Symbols xii List of Abbreviations xviList of Figures 3.1 Modelling of a simple navigation task with an MDP. 3.2 (a) The forward and (b) backward triggered observation model. 3.3 An example policy tree of a POMDP.3.4 The one step POMDP value function for each action. 3.5The complete one step POMDP value function. 3.6The two step policy tree for executing action a 1 . 3.7The transformed horizon 1 value function for taking action a 1 and perceiving each of the possible observations.3.8 The two step value function for executing action a 1 as the initial action. 4.1 State space hierarchy decomposition. The figure depicts the decomposition of a top level state to lower level states. The top level state corresponds to 4 POMDPs at level 2, each one decomposing the location of the top level state into 4 locations, and its orientation in one of the ranges denoted by the shaded region of the circles for each POMDP. This state decomposition continues at lower levels until the desired discretization of the environment has been reached. 4.2 Translation and rotation of the rPOMDP transition probabilities matrix. 4.3 Planning with the RN-HPOMDP. 5.1 The data set used for NN training. 5.2The results obtained from the trained NN.5.3 The``hot'' points defined for the FORTH main entrance hall, marked with``x". 5.4 An example of making long-term prediction for an object's movement.5.5 The probability assignment for possible hot points is dependent on the angular distance of the considered cell and the GDO and its distance from the obstacle's current position. 5.6The map of``hot'' points obtained for the FORTH main entrance hall. 5.7Tracking the motion of two persons. 5.8 (a) The static and (b) dynamic RGM. Reward discount is performed according to the obtained long-term prediction. Long-term predictions for hot points present in the periphery of the field-of-view have low probability, w i , and thus the reward discount is smaller. 6.1 An example policy tree of a POMDP with pairs of actions and speeds. 6.2 Definition of the projected state s p . 6.3 (a) The static RGM and (b) An example of the robot choosing to move with the f ast velocity. 103 7.1 Lefkos. 109 7.2 Evaluation of the learned RN-HPOMDP model. 111 7.3 The marked locations in the environment where the experimental evaluation of the RN-HPOMDP model was performed. 112 7.4 Avoiding two moving objects with a detour (I). 114 7.5 Avoiding two moving objects with a detour (II). 115 7.6 Deciding to follow a completely different path (I). 7.7 Deciding to follow a completely different path (II).7.8 Avoiding obstacles by decreasing the robot's speed.7.9 Avoiding obstacles by increasing the robot's speed. 7.10 An example of how the human motion areas are defined for the comparative experiments performed.IDC Iterative Dual Correspondence algorithm for scan mathcing.
This paper considers the problem of autonomous robot navigation in dynamic and congested environments. The predictive navigation paradigm is proposed where probabilistic planning is integrated with obstacle avoidance along with future motion prediction of humans and/or other obstacles. Predictive navigation is performed in a global manner with the use of a hierarchical Partially Observable Markov Decision Process (POMDP) that can be solved online at each time step and provides the actual actions the robot performs. Obstacle avoidance is performed within the predictive navigation model with a novel approach by deciding paths to the goal position that are not obstructed by other moving objects movement with the use of future motion prediction and by enabling the robot to increase or decrease its speed of movement or by performing detours. The robot is able to decide which obstacle avoidance behavior is optimal in each case within the unified navigation model employed.
This paper proposes a new hierarchical formulation of POMDPs for autonomous robot navigation that can be solved in real-time, and is memory efficient. It will be referred to in this paper as the Robot Navigation-Hierarchical POMDP (RN-HPOMDP). The RN-HPOMDP is utilized as a unified framework for autonomous robot navigation in dynamic environments. As such, it is used for localization, planning and local obstacle avoidance. Hence, the RN-HPOMDP decides at each time step the actions the robot should execute, without the intervention of any other external module for obstacle avoidance or localization. Our approach employs state space and action space hierarchy, and can effectively model large environments at a fine resolution. Finally, the notion of the reference POMDP is introduced. The latter holds all the information regarding motion and sensor uncertainty, which makes the proposed hierarchical structure memory efficient and enables fast learning. The RN-HPOMDP has been experimentally validated in real dynamic environments.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.