
This paper introduces TransDreamer, a novel reinforcement learning agent that leverages transformers for improved long-term memory and reasoning in visual control tasks, outperforming the previous state-of-the-art, Dreamer, in complex environments requiring long-range dependencies.


coremsg

transdreamer-a-transformer-based-reinforcement-learning-agent-with-improved-long-term-memory-for-visual-control-tasks


TransDreamer: A Transformer-Based Reinforcement Learning Agent with Improved Long-Term Memory for Visual Control Tasks


title_rewrite


QT-TDM, a novel model-based reinforcement learning algorithm, leverages the strengths of Transformer Dynamics Models (TDM) and Autoregressive Q-Learning to achieve superior performance and sample efficiency in real-time continuous control tasks, effectively addressing the limitations of slow inference speed often associated with TDMs.


qt-tdm-enhancing-real-time-planning-in-reinforcement-learning-by-combining-transformer-dynamics-model-and-autoregressive-q-learning-for-improved-speed-and-performance


QT-TDM: Enhancing Real-Time Planning in Reinforcement Learning by Combining Transformer Dynamics Model and Autoregressive Q-Learning for Improved Speed and Performance
