DD-PPO: Learning Near-Perfect PointGoal Navigators from 2.5 Billion Frames

Wijmans, Erik; Kadian, Abhishek; Morcos, Ari; Lee, Stefan; Essa, Irfan; Parikh, Devi; Savva, Manolis; Batra, Dhruv

doi:10.48550/arxiv.1911.00357

Cited by 26 publications

(50 citation statements)

References 31 publications

(70 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…r t encourages the agent to take actions that help maximize the next separation at each step. We train the policy using Decentralized Distributed PPO (DD-PPO) [75] with trajectory rollouts of 20 steps. The DD-PPO loss consists of a value network loss, policy network loss, and entropy loss to promote exploration (see Supp.…”

Section: Audio-visual Motion Policymentioning

confidence: 99%

Active Audio-Visual Separation of Dynamic Sound Sources

Majumder¹,

Al-Halah²,

Grauman³

2022

Preprint

View full text Add to dashboard Cite

We explore active audio-visual separation for dynamic sound sources, where an embodied agent moves intelligently in a 3D environment to continuously isolate the time-varying audio stream being emitted by an object of interest. The agent hears a mixed stream of multiple audio sources (e.g., multiple people conversing and a band playing music at a noisy party). Given a limited time budget, it needs to extract the target sound using egocentric audio-visual observations. We propose a reinforcement learning agent equipped with a novel transformer memory that learns motion policies to control its camera and microphone to recover the dynamic target audio, improving its own estimates for past timesteps via self-attention. Using highly realistic acoustic SoundSpaces [14] simulations in real-world scanned Mat-terport3D [12] environments, we show that our model is able to learn efficient behavior to carry out continuous separation of a time-varying audio target. Project: https: //vision.cs.utexas.edu/projects/activeav-dynamic-separation/.

show abstract

Section: Audio-visual Motion Policymentioning

confidence: 99%

Active Audio-Visual Separation of Dynamic Sound Sources

Majumder¹,

Al-Halah²,

Grauman³

2022

Preprint

View full text Add to dashboard Cite

show abstract

“…Robotic navigation. Robots often use visual signals to navigate in novel environments [64,65,66,67]. While vision is often a reliable cue for depth estimation, there are many situations where it is unavailable (e.g.…”

Section: Static Motionmentioning

confidence: 99%

Structure from Silence: Learning Scene Structure from Ambient Sound

Chen¹,

Hu²,

Owens³

2021

Preprint

View full text Add to dashboard Cite

https://ificl.github.io/structure-from-silence (a) Quiet Campus dataset (b) Depth estimation (c) Multimodal self-supervision ( ) , ( ) , (a) Quiet Campus dataset (b) Depth estimation (c) Multimodal self-supervision Figure 1: What can ambient sound tell us about 3D scene structure? (a) We collect an "in-the-wild" dataset of paired audio and RGB-D recordings from quiet indoor scenes. (b) Given audio from a scene, we estimate distance to a wall. (c) We use this ambient sound to learn audio-visual representations through self-supervision.

show abstract

“…Although in the machine learning community there have been several works related to visual navigation in either game-like domains [16], [10] or robotic environments [17], [18], the robotics community has been reluctant to adopt these systems, believing classic methods to perform better for generic tasks [19]. In this context, [5] shows that with sufficient training, RL-based methods can outperform classic methods on fair settings, and [20] shows that RL agents can solve visual navigation tasks with an almost perfect SPL score [15] in simulations. We believe this to be the right direction, but there are still several limitations that need to be addressed, as we show in Section IV.…”

Section: Related Work a Visual Navigationmentioning

confidence: 99%

Point Cloud Based Reinforcement Learning for Sim-to-Real and Partial Observability in Visual Navigation

Lobos-Tsunekawa

Harada

2020

Preprint

View full text Add to dashboard Cite

Reinforcement Learning (RL), among other learning-based methods, represents powerful tools to solve complex robotic tasks (e.g., actuation, manipulation, navigation, etc.), with the need for real-world data to train these systems as one of its most important limitations. The use of simulators is one way to address this issue, yet knowledge acquired in simulations does not work directly in the real-world, which is known as the sim-to-real transfer problem. While previous works focus on the nature of the images used as observations (e.g., textures and lighting), which has proven useful for a sim-to-sim transfer, they neglect other concerns regarding said observations, such as precise geometrical meanings, failing at robot-to-robot, and thus in sim-to-real transfers. We propose a method that learns on an observation space constructed by point clouds and environment randomization, generalizing among robots and simulators to achieve sim-to-real, while also addressing partial observability. We demonstrate the benefits of our methodology on the point goal navigation task, in which our method proves to be highly unaffected to unseen scenarios produced by robot-to-robot transfer, outperforms image-based baselines in robot-randomized experiments, and presents high performances in sim-to-sim conditions. Finally, we perform several experiments to validate the sim-to-real transfer to a physical domestic robot platform, confirming the out-of-the-box performance of our system.

show abstract

DD-PPO: Learning Near-Perfect PointGoal Navigators from 2.5 Billion Frames

Cited by 26 publications

References 31 publications

Active Audio-Visual Separation of Dynamic Sound Sources

Active Audio-Visual Separation of Dynamic Sound Sources

Structure from Silence: Learning Scene Structure from Ambient Sound

Point Cloud Based Reinforcement Learning for Sim-to-Real and Partial Observability in Visual Navigation

Contact Info

Product

Resources

About