Abstract:

In this paper, the problem of safe motion planning of the mobile agent in the presence of multiple static obstacles is investigated. In addition to the collision avoidance, an additional objective of joint minimizing the energy consumption for con- trolling its dynamic movement and maximizing the instantaneous post-processing signal-to-noise ratio (ISNR) that determines the accessing capability of spectrum allocated to the licensed users is taken into account. Due to a non-existing system setup for single carrier transmissions in the spatial-temporal correlated frequency selective fading channel and non-existing feasible mathematical analysis to maximize the distribution of the ISNR over the energy conscious motion planning, we propose a model- free and off-policy soft actor critic (SAC) to learn and make an action by the mobile agent to achieve the following three objectives simultaneously from interactions with the environment: i) achieve the safe motion planning that avoids collision with the static obstacles; ii) minimize the control energy consumption; and iii) maximize the ISNR. Simulation results verify that these three objectives can be achieved efficiently by the proposed SAC-based safe motion planning.