Driving · 2026-10-14
Axiom-0: Hierarchical Self-Refinement for Vision-Language-Action Policies
Renbin Shu, Jing Yang, Xiangyuan You, Guolin Qiu, Yufei Xia, Yiming Zhu, Nanjun Dai, Fei Zhang, Feng Wu, Keke Tang
AI-assisted summary based on the paper. Consult the linked source for methods and limitations. Benchmark numbers appear after editor verification.
TL;DR
Axiom-0 is a VLA (vision-language-action) policy that uses hierarchical self-refinement. It adds IVRR to supervise selected intermediate VLM features with learned projections aligned to spatial foundation-model references, and VAIR to recursively refine the VLM conditioning before action generation in a VLM–DiT policy. On RoboCasa365, the authors report 62.32% overall success.
Why it matters
The authors address limits of VLA models under unfamiliar task compositions by coupling representation quality with how it conditions action generation. They report improved RoboCasa365 task completion via a refine-before-act hierarchy (IVRR + VAIR) with a stated small increase in mean action-chunk generation latency under their profiling setup.
Method
- IVRR: supervise selected intermediate VLM features using learned projections aligned to spatial foundation-model reference features during training; deployment uses learned VLM features without the auxiliary alignment branch.
- VAIR: recursively refine conditioning values at adapted VLM interfaces using a shared recurrent Transformer block with gated residual updates before DiT action sampling.
- Two-stage training: Stage I optimizes VLM+DiT with alignment supervision; Stage II jointly optimizes supervised VLM layers, VAIR, and DiT under the action objective; at inference only refined conditioning is computed.
Limitation
The authors state: “Evaluation and comparison scope. Our success rates are evaluated by the authors and are pending independent benchmark verification.” They also note profiling is limited to “one device and workload” and that “warm-up counts, numerical precision, and detailed timing boundaries are not recorded in the available measurements.”
Abstract (from arXiv)
Reliable manipulation requires policies that generalize across tasks, layouts, and object interactions. Vision-language-action (VLA) models combine pretrained multimodal understanding with learned control, but their performance remains limited on unfamiliar task compositions. We present Axiom-0, a VLA policy that addresses both the quality of its representations and their use in action generation through hierarchical self-refinement. Intra-VLM Representation Refinement (IVRR) strengthens selected intermediate features through consistency supervision with learned feature projections during training. VLM-Action Interface Refinement (VAIR) recursively updates the resulting conditioning before action generation. Joint optimization of the supervised VLM layers, VAIR, and the action generator connects the two components. On RoboCasa365, Axiom-0 achieves 62.32% overall success in our evaluation, exceeding the highest published leaderboard score (Paimon-0, 58.1%) by 4.22 percentage points and the Xiaomi-Robotics-1 reference by 4.92 points. Under the evaluated profiling setup, mean action-chunk generation latency increases by 1.7% over Xiaomi-Robotics-1. These results demonstrate improved task completion with modest inference overhead.
Related papers
- Causal Analysis of Robot Foundation Models Enables Efficiency Gains
- Fast-VLA: You Only Think Once for Robot Manipulation
- Rewiring Semantics, Dynamics, and Control: A Simple yet Effective Action-Centric Tri-Stream Transformer
- PaceVLA: Adaptive Temporal Action Quantization in Vision-Language-Action Models
- MIKASA-Robo-VLA: Benchmarking Memory in VLA Models for Long-Horizon Manipulation
Publication record
Bibliographic evidence describes where and when a record appeared. It does not validate scientific claims.
First public release: 2026-10-09 · Indexed by Pub-AI: 2026-10-14 UTC
Axiom-0: Hierarchical Self-Refinement for Vision-Language-Action Policies
Preprint posted: 2026-10-09 · Original record ↗
- arxiv
- 2610.14155
Explicit source versions (1)
- v1 · 2026-10-09
- Preprint available
Source date: 2026-10-09 · First detected: 2026-10-25 UTC · Historical baseline
Source evidence (1)
- Official evidence ↗ · atom:published (official metadata)
Abstract · arxiv
Reliable manipulation requires policies that generalize across tasks, layouts, and object interactions. Vision-language-action (VLA) models combine pretrained multimodal understanding with learned control, but their performance remains limited on unfamiliar task compositions. We present Axiom-0, a VLA policy that addresses both the quality of its representations and their use in action generation through hierarchical self-refinement. Intra-VLM Representation Refinement (IVRR) strengthens selected intermediate features through consistency supervision with learned feature projections during training. VLM-Action Interface Refinement (VAIR) recursively updates the resulting conditioning before action generation. Joint optimization of the supervised VLM layers, VAIR, and the action generator connects the two components. On RoboCasa365, Axiom-0 achieves 62.32% overall success in our evaluation, exceeding the highest published leaderboard score (Paimon-0, 58.1%) by 4.22 percentage points and the Xiaomi-Robotics-1 reference by 4.92 points. Under the evaluated profiling setup, mean action-chunk generation latency increases by 1.7% over Xiaomi-Robotics-1. These results demonstrate improved task completion with modest inference overhead.