Training Agent for First-Person Shooter Game with Actor-Critic Curriculum Learning

By Yuxin Wu, Yuandong Tian. 2016

196 citations recorded by the source corpus.

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

Citations in the corpus

Learning World Graphs to Accelerate Hierarchical Reinforcement Learning (2019, positive): “A2C’s hierarchical extension FN consists of a high-level controller (“Manager”), which learns to propose subgoals to the low-level controller (“Worker”), which learns to complete the subgoals.”
Discovering Diverse Multi-Agent Strategic Behavior via Reward Randomization (2021, background): “…feature vector φ ( s, a ) can be written as a 2-dimensional vector [ (cid:107) L T − L A (cid:107) 2 , I ( L T = L A )] ; in real-time strategy games (Wu & Tian, 2016; Vinyals et al., 2017; OpenAI et al., 2019), φ is typically related to the bonus points for destroying each type of units; in…”
From Few to More: Large-scale Dynamic Multiagent Curriculum Learning (2019, background): “Later, Wu and Tian (2017) integrated RL with CL for the complex video game Doom, and developed an adaptive curriculum training that samples from a varying distribution of tasks to train the model, which achieves higher scores than learning the target task directly.”
MentorNet: Regularizing Very Deep Neural Networks on Corrupted Labels (2017, background): “find better minima in a variety of computer vision problems [38, 26, 7, 16, 25, 44], face detection [26], object detection [7], video detection [16].”
Prosocial learning agents solve generalized Stag Hunts better than selfish ones (2017, background): “Reactive training works very well in zero-sum situations, where for one agent to gain the other must lose (Silver et al. 2016; Wu and Tian 2016).”
Influence-Based Multi-Agent Exploration (2019, background): “…learning rely on a dense shaped reward function, such as distance to the goal (Mirowski et al., 2016; Wu et al., 2018), scores in games (Mnih et al., 2015) or expert-designed rewards (Wu & Tian, 2016; OpenAI, 2018), while tend to struggle in many real-world scenarios with sparse rewards.”
Advantage Actor-Critic Methods for CarRacing (2018, background): “In recent years, advantage actor-critic methods have achieved many impressive results, from visual navigation (Zhu et al., 2017) to game playing (Y. Wu & Tian, 2016).”
Synthetic Data for Deep Learning (2019, background): “[314] and later used in many important works [15,343,480,648].”
StarCraft Micromanagement With Reinforcement Learning and Curriculum Transfer Learning (2018, methodology): “learning and transfer learning, curriculum transfer learning (CTL) method has shown good performance to help the learning process converge faster and towards better optimum in recent work [54], [55], [56].”
Virtual-to-Real: Learning to Control in Visual Semantic Segmentation (2018, background): “, 2016], has been proven data-efficient and suitable for vision based control tasks [Wu and Tian, 2016; Parisotto and Salakhutdinov, 2017; You et al., 2017; Zhu et al., 2017].”
TStarBots: Defeating the Cheating Level Builtin AI in StarCraft II in the Full Game (2018, background): “, Atari [8], the first person shooting game Doom/ViZDoom [9, 10] or Quake/DeepmindLab [11], Dota 2 [12]), Robot Visuomotor Control [13], Robot Navigation [14, 15], etc.”
An Introduction to Deep Reinforcement Learning (2018, unknown): “VizDoom has been used as a platform for investigations of reward shaping (Lample and Chaplot, 2017), curriculum learning (Wu and Tian, 2016), predictive planning (Dosovitskiy and Koltun, 2016), and meta-reinforcement learning (Duan et al., 2016b).”
Actor-Critic Policy Optimization in Partially Observable Multiagent Environments (2018, background): “These techniques have learned to play Atari games beyond human-level [60], Go, chess, and shogi from scratch [82, 81], complex behaviors in 3D environments [59, 97, 37], robotics [27, 73], character animation [67], among others.”