Evo-Depth: A Lightweight Depth-Enhanced Vision-Language-Action Model

2026-05-14

AI Summary

This work addresses the limitations of existing vision-language-action models, which often rely on 2D representations lacking depth cues and thus struggle with tasks requiring precise spatial understanding. While explicitly incorporating 3D inputs can improve performance, it typically increases system complexity or necessitates additional sensors. To overcome this, the authors propose a lightweight depth-enhanced framework that leverages only multi-view RGB images. Their approach employs an implicit depth encoding module to extract compact depth features, which are then seamlessly integrated into vision-language representations through depth-aware modulation and spatial enhancement mechanisms. Coupled with a progressive alignment training strategy, the method significantly boosts spatial reasoning and action generation capabilities while maintaining low computational overhead. Experiments demonstrate state-of-the-art performance across four simulation benchmarks and the highest average success rate on real-world robotic tasks, achieving the smallest model size, lowest GPU memory consumption, and fastest inference speed.

Abstract

Vision-Language-Action models have emerged as a promising paradigm for robotic manipulation by unifying perception, language grounding, and action generation. However, they often struggle in scenarios requiring precise spatial understanding, as current VLA models primarily rely on 2D visual representations that lack depth information and detailed spatial relationships. While recent approaches incorporate explicit 3D inputs such as depth maps or point clouds to address this issue, they often increase system complexity, require additional sensors, and remain vulnerable to sensing noise and reconstruction errors. Another line of work explores implicit 3D-aware spatial modeling directly from RGB observations without extra sensors, but it often relies on large geometry foundation models, resulting in higher training and deployment costs. To address these challenges, we propose Evo-Depth, a lightweight depth-enhanced VLA framework that enhances spatially grounded manipulation without relying on additional sensing hardware or compromising deployment efficiency. Evo-Depth employs a lightweight Implicit Depth Encoding Module to extract compact depth features from multi-view RGB images. These features are incorporated into vision-language representations through a Spatial Enhancement Module via depth-aware modulation, enabling efficient spatial-semantic enhancement. A Progressive Alignment Training strategy is further introduced to align the resulting depth-enhanced representations with downstream action learning. With only 0.9B parameters, Evo-Depth achieves superior performance across four simulation benchmarks. In real-world experiments, Evo-Depth attains the highest average success rate while also exhibiting the smallest model size, lowest GPU memory usage, and highest inference frequency among compared methods.

Problem

• Vision-Language-Action models
• spatial understanding
• depth information
• 3D-aware modeling
• robotic manipulation

Innovation

• Implicit Depth Encoding
• Spatial Enhancement
• Vision-Language-Action Model
• Lightweight Robotics
• Progressive Alignment Training