# InternVLA-A1.5: Unifying Understanding, Latent Foresight, and Action for Compositional Generalization

arXiv: https://arxiv.org/abs/2607.04988 · HTML: https://mlsanity.com/paper/2607.04988/

## Key facts

- Cited by 20 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 6 Jul 2026.

## Built on by (most cited first, 20 of 20)

- [Data Pyramid for Embodied Manipulation: A Survey](https://mlsanity.com/paper/2607.24744/) (2026)
- [Beyond Data Scaling: Representation-Centric Continued Pre-training for Vision-Language-Action Models](https://mlsanity.com/paper/2608.27550/) (2026)
- [DynamicWAM: Dual-Path Motion Conditioning for World-Action Models in Dynamic Manipulation](https://mlsanity.com/paper/2608.00793/) (2026)
- [ZimaBlue: Evolving Generalizable World Action Models through Scalable Video Pre-training](https://mlsanity.com/paper/2609.00188/) (2026)
- [N0-TWAM: Scaling Tactile-Native World-Action Model for Contact-Rich Manipulation](https://mlsanity.com/paper/2607.23783/) (2026)
- [Rolling-WAM: World Action Models with Rolling Imagination](https://mlsanity.com/paper/2609.30247/) (2026)
- [InternW0-Δ: A World Action Model Bridging Predictive Dynamics and Actions with 20K+ Hours of Open Data](https://mlsanity.com/paper/2609.31394/) (2026)
- [Embodied Turing Machines: Stateful Code for Robot Recursive Self-Improvement](https://mlsanity.com/paper/2610.12369/) (2026)
- [RoboInter1.5: A Holistic Intermediate Representation Suite for Embodied World Modeling and Robotic Manipulation](https://mlsanity.com/paper/2607.18709/) (2026)
- [LIBERO-VIFO: Benchmarking the Capability and Safety of Visual Cue Following in Vision-Language-Action Models](https://mlsanity.com/paper/2608.17600/) (2026)
- [PAVE: Predictive Alignment and Value-Guided Evolution for World-Action Policies](https://mlsanity.com/paper/2608.30378/) (2026)
- [An Action Is Worth One Patch: Unified World-Action Modeling with PatchWAM](https://mlsanity.com/paper/2609.25961/) (2026)
- [Quantile Head for Vision-Language-Action Models](https://mlsanity.com/paper/2609.34061/) (2026)
- [Staircase Policy: Streaming Inference for World-Action Models with Large Action Chunks](https://mlsanity.com/paper/2609.36471/) (2026)
- [CogWAM: Aligning Semantic Cognition with World Action Modeling via Event-Driven Interfaces](https://mlsanity.com/paper/2609.37721/) (2026)
- [Vela: Scaling Vision-Language-Action Models with Adaptive Action Curve Parametrization](https://mlsanity.com/paper/2610.05230/) (2026)
- [Juno: Taming Predictive Latents for Vision-Language-Action Models](https://mlsanity.com/paper/2610.09940/) (2026)
- [Rewiring Semantics, Dynamics, and Control: A Simple yet Effective Action-Centric Tri-Stream Transformer](https://mlsanity.com/paper/2610.11416/) (2026)
- [RoboAware: Learning to Coordinate Embodied Skills from Counterfactual Outcomes](https://mlsanity.com/paper/2610.11480/) (2026)
- [What Matters for World Modeling in Vision-Language-Action Models?](https://mlsanity.com/paper/2610.19825/) (2026)

## Abstract (arXiv)

Unified models for robot manipulation aim to equip one policy with both the semantic priors of pretrained VLMs and the physical dynamics learned through future prediction. In practice, existing designs tend to erode the semantics of the pretrained backbone, suffer interference among heterogeneous objectives, and learn future prediction from scratch in pixel space, leaving the dynamics priors of pretrained video generators unexploited. We present InternVLA-A1.5, which builds the policy on a native VLM backbone that keeps training on VQA and subtask prediction, and attaches a lightweight unified expert for continuous action generation. Future prediction is recast as a latent-querying problem, where a small set of learnable foresight tokens condenses the task-relevant future into a compact latent code under the supervision of a frozen pretrained video generation model, so the policy inherits world-model dynamics priors without ever learning pixel-level generation. The video branch is discarded at inference, keeping real-time control. Pretrained on 1.2M robot episodes and 3M multimodal samples, InternVLA-A1.5 achieves the best overall results on all six simulation benchmarks. In the real world, the preserved semantics deliver the strongest compositional generalization on held-out instruction bindings, and the two designs together sustain long-horizon execution.