|
G-K Curve-based Knee Point Prediction Method for Li-ion Batteries |
|
Kwangrae Kim, Minho Kim, Huiyong Chun(POSTECH, Korea), Kyunghwan Lee(Samsung SDI Co., LTD., Korea), Soohee Han(POSTECH, Korea) |
| Lithium-ion batteries (LIBs) are very promising energy storage devices, and predicting the pattern of capacity fade is very important for both manufacturers and users. In particular, with the recent surge in interest in battery reuse, a method that can predict the ’knee point’ phenomenon, which is a rapid decrease in capacity occurring in the later stage of battery life, is becoming more critical. This paper proposes a Gradient 1-Knee point (G-K) curve that can predict the knee point simply using long-term experimental data of Samsung SDI commercial cell (2170 NMC). The proposed G-K curve allows us to estimate the approximate knee point even on practical devices with limited computing power. |
|
|
Invariance Domain for the L1 Performance of Nonlinear Systems |
|
Hyung Tae Choi(POSTECH, Korea), Jung Hoon Kim(Pohang University of Science and Technology, Korea) |
| This paper aims at characterizing the L1 performance of nonlinear systems, by which we mean the maximum ratio of L infinity norms of input and output signals. The set-invariance methods such as an invariant set and an invariance domain have been used for the characterization. More precisely, a sufficient condition for the L1 performance is derived in terms of the invariant set. Also, the other sufficient condition is constructed in terms of an extended version of the conventional invariance domain. Both results do not assume the uniqueness of solutions for the differential equations. |
|
|
Automatic Data Augmentation by Upper Confidence Bound for Deep Reinforcement Learning |
|
Soohee Han, Yoonhee Gil, Jongchan Baek, Jonghyuk Park(POSTECH, Korea) |
| In this paper, we suggest a new automatic data augmentation method that can improve generalization capabilities of visual RL agent. Among various kinds of data augmentations, this method chooses the best augmentation method for RL agent by upper confidence bounds (UCB). This method can be applied to any RL algorithms that implements data augmentations. In the experiment, the proposed method showed improved generalization capabilities in DMControl Generalization Benchmark (DMControl-GB), a modified DeepMind Control Suite (DMControl) that can randomize the color of the agent or background. |
|
|
Event-triggered Actor Critic through Entropy Regularization for Highly Sparse Robot Controller |
|
JongHyuk Park, Soohee Han(POSTECH, Korea) |
| This paper proposes event-triggered actor critic, which can learn a highly sparse robot controller via reinforcement learning. |
|
|
Stability Analysis of Time-varying Delay Neural Network System Utilizing Free-matrix-based Double Integral Inequality |
|
Jun Hui Lee, Hyeon-Woo Na, Poogyeon Park(POSTECH, Korea) |
| This paper proposes two free-matrix-based double integral inequality. Unlike the generally widely used double integral inequalities, it provides a convex-type estimation and guarantees a tighter upper bound than the existing integral inequality by using information about the state as well as the state derivative. In addition, a less conservative stability criterion on time-varying delay neural network system is derived using the proposed double integral inequality. We evaluate the superiority of the proposed free matrix-based double integral inequality through well-known numerical example. |
|
|
Energy Consumption Analysis of a Downward Tethered quadrotor |
|
ChangHyeon Lee(POSTECH, Korea), Junwoo Jason Son(Pohang University of Science and Technology, Korea), Hakjun Lee(Polaris 3D, Korea), Soohee Han(Pohang University of Science and Technology, Korea) |
| This paper analyze the energy consumption of a new concept of tethered quadrotor system, downward tethered quadrotor (DTQ), to provide insight about using DTQ. DTQ is one of a tethered quadrotor system in which station is located above the flight level of the quadrotor. Thanks to its system layout, the tether can play an important role in energy efficient flight of the quadrotor by adjusting the tension properly. In order to extremize the advantage of DTQ, optimization problem for mechanical power consumption of DTQ in hovering states is formulated. |
|