EvoVLA: Self-Evolving Vision-Language-Action Model

TL;DR

EvoVLA introduces a self-supervised framework for long-horizon robotic manipulation that reduces stage hallucination, improves success rates, and enhances sim-to-real transfer through novel components like SAR, POE, and Long-Horizon Memory.

Contribution

EvoVLA presents a new self-supervised VLA model with three key components to address stage hallucination and improve long-horizon manipulation performance.

Findings

Achieves 10.2% higher success rate than baseline on Discoverse-L.

Reduces stage hallucination from 38.5% to 14.8%.

Outperforms baseline in real-world robot experiments with 54.6% success.

Abstract

Long-horizon robotic manipulation remains challenging for Vision-Language-Action (VLA) models despite recent progress in zero-shot generalization and simulation-to-real-world transfer. Current VLA models suffer from stage hallucination, where agents exploit coarse evaluation signals to shortcut multi-step tasks, reporting high progress without truly completing them. We present EvoVLA, a self-supervised VLA framework that addresses this issue through three complementary components: Stage-Aligned Reward (SAR), which uses triplet contrastive learning with Gemini-generated hard negatives to prevent visual shortcuts; Pose-Based Object Exploration (POE), which grounds curiosity in relative object-gripper pose instead of raw pixels; and Long-Horizon Memory, which uses selective context retention and gated fusion to stabilize intrinsic shaping during extended rollouts. Extensive evaluations on Discoverse-L, a long-horizon manipulation benchmark with three multi-stage tasks, show that EvoVLA improves average task success by 10.2 percentage points over the strongest baseline (OpenVLA-OFT), reaching 69.2 percent. EvoVLA also achieves one-and-a-half times better sample efficiency and reduces stage hallucination from 38.5 percent to 14.8 percent. Real-world deployment on physical robots reaches an average success rate of 54.6 percent across four manipulation tasks, outperforming OpenVLA-OFT by 11 points, demonstrating effective sim-to-real transfer and strong generalization. Code: https://github.com/AIGeeksGroup/EvoVLA. Website: https://aigeeksgroup.github.io/EvoVLA.