Act2Goal: From World Model To General Goal-conditioned Policy

TL;DR

Act2Goal introduces a goal-conditioned manipulation policy combining a visual world model with multi-scale temporal control, enabling robust, long-horizon robotic tasks with strong zero-shot generalization and rapid online adaptation.

Contribution

The paper presents Act2Goal, a novel approach integrating a visual world model with multi-scale temporal hashing for improved long-horizon manipulation.

Findings

Achieves success rates from 30% to 90% on out-of-distribution tasks.

Enables rapid autonomous improvement through reward-free online adaptation.

Demonstrates strong zero-shot generalization to new objects and environments.

Abstract

Specifying robotic manipulation tasks in a manner that is both expressive and precise remains a central challenge. While visual goals provide a compact and unambiguous task specification, existing goal-conditioned policies often struggle with long-horizon manipulation due to their reliance on single-step action prediction without explicit modeling of task progress. We propose Act2Goal, a general goal-conditioned manipulation policy that integrates a goal-conditioned visual world model with multi-scale temporal control. Given a current observation and a target visual goal, the world model generates a plausible sequence of intermediate visual states that captures long-horizon structure. To translate this visual plan into robust execution, we introduce Multi-Scale Temporal Hashing (MSTH), which decomposes the imagined trajectory into dense proximal frames for fine-grained closed-loop control and sparse distal frames that anchor global task consistency. The policy couples these representations with motor control through end-to-end cross-attention, enabling coherent long-horizon behavior while remaining reactive to local disturbances. Act2Goal achieves strong zero-shot generalization to novel objects, spatial layouts, and environments. We further enable reward-free online adaptation through hindsight goal relabeling with LoRA-based finetuning, allowing rapid autonomous improvement without external supervision. Real-robot experiments demonstrate that Act2Goal improves success rates from 30% to 90% on challenging out-of-distribution tasks within minutes of autonomous interaction, validating that goal-conditioned world models with multi-scale temporal control provide structured guidance necessary for robust long-horizon manipulation. Project page: https://act2goal.github.io/