Newest papers

VaDar: Implicit Goal Conditioning via Value Disaggregation

Shashwat Saxena, Mehul Goel, Sreyas Venkataraman, Sarvesh Patil, Max Simchowitz · 2026 · arXiv (Cornell University)

DOI: 10.48550/arxiv.2610.26629 · cited by 0 works (OpenAlex) · open-access PDF

Portuguese

Tarefas de longo prazo são importantes para a IA moderna e a robótica, mas representam um desafio para a aprendizagem de reforço. Historicamente, uma abordagem popular tem sido a decomposição de uma tarefa em sub-objetivos e a aprendizagem de uma política condicionada por sub-objetivo para executar cada sub-tarefa em sequência. Introduzimos uma alternativa chamada Value Disaggregation (VaDar), que treina uma única política independente de sub-objetivo para maximizar uma combinação linear de valores críticos por sub-objetivo. Porque a informação sub-objetivo é necessária apenas no tempo de treinamento, o VaDar elimina a necessidade de uma seleção sub-objetiva possivelmente dispendiosa durante a implementação. Além disso, através de experiências cuidadosas, most

Machine translation (Meta M2M100); the English original follows.

Abstract

Long-horizon tasks are important for modern AI and robotics, yet pose a challenge for reinforcement learning. Historically, a popular approach has been to decompose a task into subgoals, and learn a subgoal-conditioned policy to execute each subtask in sequence. We introduce an alternative called Value Disaggregation (VaDar), which trains a single subgoal-independent policy to maximize a linear combination of per-subgoal critic values. Because subgoal information is needed only at training-time, VaDar removes the need for possibly costly or cumbersome subgoal-selection during policy deployment. Moreover, through careful experiments, we show that VaDar (1) is always competitive with, and frequently outperforms, natural baselines like goal-conditioning and (2) succeeds on tasks with non-sequential goal structure and multiple optimization objectives. In particular, VaDar is the first algorithm to solve challenging, long-horizon tasks in the RoboCasa suite, even when starting from a policy pretrained from behavior cloning with near-zero initial task success. Taken together, VaDar's successes suggest that subgoal information benefits actor-critic reinforcement learning by enhancing the efficacy of critic learning, whereas policy goal-conditioning is often unnecessary.

Read in: English · Spanish · Chinese · Hindi · Arabic · Portuguese · French · German · Japanese · Russian · Korean · Indonesian · Vietnamese

Topics: Reinforcement learning, Task (project management), Computer science, Artificial intelligence, Value (mathematics), Machine learning

Sources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange