Browsing by Author "Kim, Jongyun"

Now showing 1 - 2 of 2
  • Thumbnail Image
    ItemOpen Access
    Receding horizon-based infotaxis with random sampling for source search and estimation in complex environments
    (IEEE, 2022-06-21) Park, Minkyu; Ladosz, Pawel; Kim, Jongyun; Oh, Hyondong
    This paper proposes a receding horizon-based information-theoretic source search and estimation strategy for a mobile sensor in an urban environment in which an invisible harmful substance is released into the atmosphere. The mobile sensor estimates the source term including its location and release rate by using sensor observations based on Bayesian inference. The sampling-based sequential Monte Carlo method, particle filter, is employed to estimate the source term state in a highly nonlinear and stochastic system. Infotaxis, the information-theoretic gradient-free search strategy is modified to find the optimal search path that maximizes the reduction of the entropy of the source term distribution. In particular, receding horizon Infotaxis is introduced to avoid falling into the local optima and to find more successful information gathering paths in obstacle-rich urban environments. Besides, a random sampling method is introduced to reduce the computational load of the receding horizon Infotaxis for real-time computation. The random sampling method samples the predicted future measurements based on current estimation of the source term and computes the optimal search path using sampled measurements rather than considering all possible future measurements. To demonstrate the benefit of the proposed approach, comprehensive numerical simulations are performed for various conditions. The proposed algorithm increases the success rate by about 30% and reduces the mean search time by about 40% compared with the existing information-theoretic search strategy.
  • Thumbnail Image
    ItemOpen Access
    Towards monocular vision-based autonomous flight through deep reinforcement learning
    (Elsevier, 2022-03-09) Kim, Minwoo; Kim, Jongyun; Jung, Minjae; Oh, Hyondong
    This paper proposes an obstacle avoidance strategy for small multi-rotor drones with a monocular camera using deep reinforcement learning. The proposed method is composed of two steps: depth estimation and navigation decision making. For the depth estimation step, a pre-trained depth estimation algorithm based on the convolutional neural network is used. On the navigation decision making step, a dueling double deep Q-network is employed with a well-designed reward function. The network is trained using the robot operating system and Gazebo simulation environment. To validate the performance and robustness of the proposed approach, simulations and real experiments have been carried out using a Parrot Bebop2 drone in various complex indoor environments. We demonstrate that the proposed algorithm successfully travels along the narrow corridors with the texture free walls, people, and boxes.