ViZDoom: A Doom-based AI Research Platform for Visual Reinforcement Learning

Kempka, Michał; Wydmuch, Marek; Runc, Grzegorz; Toczek, Jakub; Jaśkowski, Wojciech

doi:10.48550/arxiv.1605.02097

Cited by 30 publications

(51 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…It uses Tensorflow 1.x APIs and is in the style of the library tf-slim and tf.contrib.layers. With TPolicies one can build policy net or value net in various architectures, ranging from a simple one in list structure (e.g., a ConvNet plus LSTM for Atari [1,21] or ViZDoom [16]) to a complicated one of general Directed Acyclic Graph (e.g., the net for SC2 full game [8], containing layers/blocks of ResNet, Transformer, Pointer Net, Gated Linear Unit, Auto-regressive Action Heads, etc.). TPolicies also provides RL related Tensorflow ops, e.g., for building policy gradient loss, for computing λ-return.…”

Section: Code Structurementioning

confidence: 99%

“…ViZDoom [16] is an AI research platform based on the FPS (First Person Shooter) game Doom. We adopt the CIG 2016 competition track 1 protocol [65], where 8 AI players join in a maze and play against each other.…”

Section: Vizdoommentioning

confidence: 99%

“…Then we describe CSP-MARL and explain the design of our code implementation in Section 3. Finally, we discuss several experiments over StarCraft 2, ViZDoom [16] and Pommerman [17] in Section 4 to show the efficiency and effectiveness of TLeague.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

TLeague: A Framework for Competitive Self-Play based Distributed Multi-Agent Reinforcement Learning

Sun,

Xiong,

Han

et al. 2020

Preprint

View full text Add to dashboard Cite

Competitive Self-Play (CSP) based Multi-Agent Reinforcement Learning (MARL) has shown phenomenal breakthroughs recently. Strong AIs are achieved for several benchmarks, including Dota 2, Glory of Kings, Quake III, StarCraft II, to name a few. Despite the success, the MARL training is extremely data thirsty, requiring typically billions of (if not trillions of) frames be seen from the environment during training in order for learning a high performance agent. This poses non-trivial difficulties for researchers or engineers and prevents the application of MARL to a broader range of real-world problems. To address this issue, in this manuscript we describe a framework, referred to as TLeague, that aims at large-scale training and implements several main-stream CSP-MARL algorithms. The training can be deployed in either a single machine or a cluster of hybrid machines (CPUs and GPUs), where the standard Kubernetes is supported in a cloud native manner. TLeague achieves a high throughput and a reasonable scale-up when performing distributed training. Thanks to the modular design, it is also easy to extend for solving other multi-agent problems or implementing and verifying MARL algorithms. We present experiments over StarCraft II, ViZDoom and Pommerman to show the efficiency and effectiveness of TLeague. The code is open-sourced and available at https://github.com/tencent-ailab/tleague_projpage

show abstract

Section: Code Structurementioning

confidence: 99%

Section: Vizdoommentioning

confidence: 99%

See 1 more Smart Citation

TLeague: A Framework for Competitive Self-Play based Distributed Multi-Agent Reinforcement Learning

Sun,

Xiong,

Han

et al. 2020

Preprint

View full text Add to dashboard Cite

show abstract

“…Following the success of deep reinforcement learning in 3D Games such as Doom (Lample & Chaplot, 2017; and DeepmindLab , there has been increased interest in using deep reinforcement learning for training embodied AI agents, which interact with a 3D environment by receiving first-person views of the environment and taking navigational actions. The simplest navigational agents learn a particular behavior such as collecting or avoiding particular objects (Kempka et al, 2016;Jaderberg et al, 2016;Mirowski et al, 2016) or playing deathmatches (Lample & Chaplot, 2017;. Subsequently, there have been efforts on training navigational agents whose behavior is conditioned on a target specified using images (Zhu et al, 2017) or coordinates (Gupta et al, 2017a;Savva et al, 2017).…”

Section: Introductionmentioning

confidence: 99%

“…We propose a novel dualattention model involving sequential Gated-and Spatial-Attention operations to perform explicit task-invariant alignment between the image representation channels and the words in the input and answer space. We create datasets and simulation scenarios for testing cross-task knowledge transfer in the Doom environment (Kempka et al, 2016) and show an absolute improvement of 43-61% on instructions and 5-26% for questions over baselines in a range of scenarios with varying difficulty. Additionally, we demonstrate that the modularity of our model allows easy addition of new objects and attributes to a trained model.…”

Section: Introductionmentioning

confidence: 99%

Embodied Multimodal Multitask Learning

Chaplot

Lee

Salakhutdinov

et al. 2020

Proceedings of the Twenty-Ninth International Joint Conference on Artificial Intelligence

View full text Add to dashboard Cite

Visually-grounded embodied language learning models have recently shown to be effective at learning multiple multimodal tasks such as following navigational instructions and answering questions. In this paper, we address two key limitations of these models, (a) the inability to transfer the grounded knowledge across different tasks and (b) the inability to transfer to new words and concepts not seen during training using only a few examples. We propose a multitask model which facilitates knowledge transfer across tasks by disentangling the knowledge of words and visual attributes in the intermediate representations. We create scenarios and datasets to quantify cross-task knowledge transfer and show that the proposed model outperforms a range of baselines in simulated 3D environments. We also show that this disentanglement of representations makes our model modular and interpretable which allows for transfer to instructions containing new concepts.

show abstract

A proximal policy optimization with curiosity algorithm for virtual drone navigation

Das,

Khan,

Paul

2024

Eng. Res. Express

View full text Add to dashboard Cite

The drone sector is witnessing a surge in demand for advanced models tailored to address critical applications such as disaster management and intelligent warehouse deliveries. Employing simulation-based experiments with virtual drone navigation is considered a best practice before deploying physical models. Nonetheless, the current state-of-the-art virtual drone navigation system lacks accuracy and introduces notable increments in simulation time. In order to mitigate these issues, this paper introduces a deep reinforcement learning-based drone agent, designed to autonomously navigate within a constrained virtual environment. The proposed drone agent utilizes realistic drone physics in order to ensure flight within the virtual environment. The work uniquely combines & optimizes both control algorithms and physical dynamics, making the model more robust and versatile than others. The integration of curiosity-driven learning with physics-based modeling potentially increases the model’s readiness for real-world application, compared to theoretical approaches. The extensive simulation results validate the remarkable speed and accuracy of the proposed scheme compared to baseline works. The trained agent exhibits strength and versatility, enabling it to deal with the numerous targets and obstacles encountered in human environments.

show abstract

ViZDoom: A Doom-based AI Research Platform for Visual Reinforcement Learning

Cited by 30 publications

References 0 publications

TLeague: A Framework for Competitive Self-Play based Distributed Multi-Agent Reinforcement Learning

TLeague: A Framework for Competitive Self-Play based Distributed Multi-Agent Reinforcement Learning

Embodied Multimodal Multitask Learning

A proximal policy optimization with curiosity algorithm for virtual drone navigation

Contact Info

Product

Resources

About