Asking for Knowledge: Training RL Agents to Query External Knowledge Using Language

no code implementations • 12 May 2022 • Iou-Jen Liu, Xingdi Yuan, Marc-Alexandre Côté, Pierre-Yves Oudeyer, Alexander G. Schwing

In order to study how agents can be taught to query external knowledge via language, we first introduce two new environments: the grid-world-based Q-BabyAI and the text-based Q-TextWorld.

Semantic Tracklets: An Object-Centric Representation for Visual Multi-Agent Reinforcement Learning

no code implementations • 6 Aug 2021 • Iou-Jen Liu, Zhongzheng Ren, Raymond A. Yeh, Alexander G. Schwing

We evaluate `semantic tracklets' on the visual multi-agent particle environment (VMPE) and on the challenging visual multi-agent GFootball environment.

Multi-agent Reinforcement Learning reinforcement-learning +1

Cooperative Exploration for Multi-Agent Deep Reinforcement Learning

no code implementations • 23 Jul 2021 • Iou-Jen Liu, Unnat Jain, Raymond A. Yeh, Alexander G. Schwing

To address this shortcoming, in this paper, we propose cooperative multi-agent exploration (CMAE): agents share a common goal while exploring.

reinforcement-learning Reinforcement Learning (RL) +2

GridToPix: Training Embodied Agents with Minimal Supervision

no code implementations • ICCV 2021 • Unnat Jain, Iou-Jen Liu, Svetlana Lazebnik, Aniruddha Kembhavi, Luca Weihs, Alexander Schwing

While deep reinforcement learning (RL) promises freedom from hand-labeled data, great successes, especially for Embodied AI, require significant work to create supervision via carefully shaped rewards.

PointGoal Navigation Reinforcement Learning (RL) +1

Coordinated Multi-Agent Exploration Using Shared Goals

no code implementations • 1 Jan 2021 • Iou-Jen Liu, Unnat Jain, Alex Schwing

Exploration is critical for good results of deep reinforcement learning algorithms and has drawn much attention.

reinforcement-learning Reinforcement Learning (RL) +2

High-Throughput Synchronous Deep RL

1 code implementation • NeurIPS 2020 • Iou-Jen Liu, Raymond A. Yeh, Alexander G. Schwing

In contrast, asynchronous methods achieve high throughput but suffer from stability issues and lower sample efficiency due to `stale policies.'

Atari Games reinforcement-learning +2

Bridging the Imitation Gap by Adaptive Insubordination

no code implementations • NeurIPS 2021 • Luca Weihs, Unnat Jain, Iou-Jen Liu, Jordi Salvador, Svetlana Lazebnik, Aniruddha Kembhavi, Alexander Schwing

However, we show that when the teaching agent makes decisions with access to privileged information that is unavailable to the student, this information is marginalized during imitation learning, resulting in an "imitation gap" and, potentially, poor results.

Imitation Learning Memorization +2

PIC: Permutation Invariant Critic for Multi-Agent Deep Reinforcement Learning

2 code implementations • 31 Oct 2019 • Iou-Jen Liu, Raymond A. Yeh, Alexander G. Schwing

Sample efficiency and scalability to a large number of agents are two important goals for multi-agent reinforcement learning systems.

Multi-agent Reinforcement Learning reinforcement-learning +1

Knowledge Flow: Improve Upon Your Teachers

no code implementations • ICLR 2019 • Iou-Jen Liu, Jian Peng, Alexander G. Schwing

A zoo of deep nets is available these days for almost any given task, and it is increasingly unclear which net to start with when addressing a new task, or which net to use as an initialization for fine-tuning a new model.

Reinforcement Learning (RL)