An Ensemble Fuzzy Approach for Inverse Reinforcement Learning

Reinforcement Learning-Based End-to-End Parking for Automatic Parking System

“…Though the final effect converges to an ideal level, it cannot be proved that it is the optimal reward setting. Accordingly, we will consider the method of inverse reinforcement learning [33,34] to optimize the reward. (2) In this paper, the reinforcement learning-based parking only has the function of reversing (e.g., “step two” in Figure 14), and it cannot automatically adjust the gear forward and backward.…”

Ensemble Reinforcement Learning: A Survey

“…A fuzzy set can affect the reward value obtained by agents by measuring dissimilarity. Pan et al proposed a dissimilarity evaluation metric for deciding the weight value of each agent's reward in ERL [43]. In this way, ERL can achieve a good training effect with fewer iterations.…”

“…To comprehensively test the effectiveness of algorithms, different models, training algorithms, and integration methods should be separately evaluated [54]. [99] Atari games DQN Chen et al [8] Atari games A3C+ Partalas et al [100] UCI machine learning repository classifier combination methods voting (V) and SMT and the forward selection (FS), selective fusion (SF) Pearce et al [101] Cart Pole control problem Q-learning with different layer NNs Dong et al [53] traffic speed dataset GRU, LSTM, MLP, RBF, LSTM-GRU-GA Pan et al [43] Maze, Mountain Car, Robotic Soccer Game Simulation counterpart Goyal et al [46] CATS (Competition on Artificial Time series) dataset LSTM, ANN, Linear regression, Random Forest, Online NN Macheng Shen and Jonathan P How [102] two-player asymmetric game single model, RNN Qingfeng Lan et al [37] Mountain Car Q-learning, Double Q-learning, Averaged Q-learning Liu et al [54] three different groups of measured wind speed data from Xinjiang wind farms Network: LSTM method, the DBN method, the ESN method; Training algorithm: SARSA Lin et al [41] Maze, soccer robot game orthogonal projection inverse reinforcement learning method (OP-IRL) Junta Wu and Huiyun Li [73] 2D Robot Arm Open Racing Car Simulator (TORCS) DDPG Yang et al [14] Dow Jones 30 constituent stocks (at 01/01/2016) PPO, A2C, DDPG Liu et al [33] UCI online data repository classifiers combination approaches majority voting (MV), weighted voting (WV), ensemble selection methods forward selection (FS) Ghosh et al [38] open source air traffic simulator PPO Jalali et al [81] GHI data sets adaptive hybrid model (AHM), hybrid feature selection method (HFS), Outlier-robust hybrid model (ORHM), novel hybrid deep neural network model (NHDNNM), OHS-LSTM Liu et al [56] data collected from a congested intersection in Changsha RNN, ENN, ESN, DBN, RBF, GRNN, MLP Jalali et al [72] two well-known open-source image datasets named as Mendely and Kaggle original version of GSK and eight powerful evolutionary algorithms including grasshopper optimization algorithm (GOA), Slime mold algorithm (SMA), genetic algorithm, gray wolf optimizer (GWO), particle swarm optimization (PSO), differential evolution (DE), biogeographybased optimization (BBO) Hassam Ullah Sheikh et al [15] Mujoco environments, Atari games TD3, SAC and REDQ Shang et al [30] actual traffic volume data of nine stations of Changsha freeway Chebnet, CNN, LSTM, DBN, RNN, ESN, multi-layer perceptron (MLP) Tan et al …”

“…So, some researchers tried to design strategies based on scenarios, which reduce the computational overhead to some extent. An fuzzifying the reward [43]. Up to now, the number of computationally cost-reducing models is still small.…”