Organizations: INFIFORCE Intelligent Technology Co., Ltd. Hangzhou, China · Huazhong University of Science and Technology, Wuhan, China · The University of Hong Kong, Hong Kong SAR, China · Tsinghua University, Beijing, China
Vision-Language-Action (VLA) models demonstrate remarkable potential for generalizable robotic manipulation. The execution of complex multi-step behaviors in VLA models can be improved by robust instruction grounding, a critical component for effective control. However, current paradigms predominantly rely on coarse, high-level task instructions during supervised fine-tuning. This instruction grounding gap leaves models without explicit intermediate guidance, leading to severe compounding errors in long-horizon tasks. Therefore, bridging this instruction gap and providing scalable post-training for VLA models is urgent. To tackle this problem, we propose \method, the first subtask-aware VLA framework integrated with a scalable offline post-training pipeline. Our framework leverages a large language model to decompose high-level demonstrations into fine-grained atomic subtasks. This approach utilizes a pretrained predictive world model to score candidate action chunks against subtask goals in the latent space, mitigating error accumulation while significantly improving long-horizon robustness. Furthermore, this approach enables highly efficient Group Relative Policy Optimization without the prohibitive expenses associated with online rollouts on physical robots. Extensive simulations validate that our AtomVLA maintains strong robustness under perturbations. When evaluated against fundamental baseline models, it achieves an average success rate of 97.0% on the LIBERO benchmark and 48.0% on the LIBERO-PRO benchmark. Finally, experiments conducted in the real world using the Galaxea R1 Lite platform confirm its broad applicability across diverse tasks, especially long-horizon tasks. All datasets, checkpoints, and code will be released to the public domain following the acceptance of this work for future research.
Figures & tables
Fig. 2: (a) Typical VLA models rely on SFT Training. (b) AtomVLA (Ours) leverages a language model for fine-grained decomposition of atomic subtask instructions and a world model for RL post-training.
Fig. 3: Training pipeline. Stage I: high-level instructions are decomposed into fine-grained atomic subtask instructions using LLM (GPT-4o). Subsequently, these subtask instructions are integrated with the original high-level instruction to guide the SFT training of the model. Stage II: A predictive latent world model evaluates candidate action rollouts to provide reward for offline post-training.
Input
LIBERO-Long
Image
80.4%
Image + High-level Task
90.0%
Image + High-level Task + Atomic Task
92.2%
TABLE I: Different Input on LIBERO-Long. Bold denotes the best performance among all methods.
Chunk Size
Spatial
Object
Goal
Long
Avg.
4
96.4%
99.6%
97.6%
94.4%
97.0%
8
95.6%
99.2%
98.4%
93.0%
96.6%
16
96.0%
99.8%
98.8%
91.6%
96.6%
32
96.0%
99.0%
98.8%
91.2%
96.3%
TABLE II: Comparison of AtomVLA on the LIBERO benchmark with different numbers of action chunk horizon. Bold denotes the best performance among all methods.
Method
w/o Pre-Training
Model Size
LIBERO-Spatial
LIBERO-Object
LIBERO-Goal
LIBERO-Long
Avg.
TraceVLA [ 33 ]
✗
7 B
84.6%
85.2%
75.1%
54.1%
74.8%
OpenVLA [ 3 ]
✓
7 B
84.7%
88.4%
79.2%
53.7%
76.5%
SpatialVLA [ 19 ]
✓
4 B
88.2%
89.9%
78.6%
55.5%
78.1%
CoT-VLA [ 34 ]
✓
7 B
87.5%
91.6%
87.6%
69.0%
83.9%
π0 [ 6 ]
✓
4 B
96.8%
98.8%
95.8%
85.2%
94.2%
ThinkAct [ 36 ]
✗
7 B
88.3%
91.4%
87.1%
70.9%
84.4%
TABLE III: Comparisons with state-of-the-art methods on the LIBERO benchmark [ 31 ] . Bold denotes the best performance among all methods.
Method
LIBERO-Goal
LIBERO-Spatial
LIBERO-10
LIBERO-Object
Avg.
Obj
Pos
Sem
Task
Obj
Pos
Sem
Task
Obj
Pos
Sem
Task
Obj
Pos
Sem
Task
π0 [ 6 ]
0.94
0.00
0.93
0.00
0.95
0.00
0.97
0.00
0.79
0.00
0.82
0.00
0.94
0.00
0.90
0.00
0.45
MolmoAct [ 37 ]
0.68
0.00
0.85
0.00
0.90
0.00
0.88
0.00
0.54
0.00
0.74
0.06
0.92
0.06
0.96
0.00
0.41
NORA [ 35 ]
0.58
0.00
0.88
0.00
0.92
0.00
0.91
0.00
0.46
0.00
0.74
0.00
0.86
0.00
0.92
0.00
0.39
X-VLA [ 38 ]
0.68
0.01
0.98
0.09
0.97
0.00
0.96
0.00
0.62
0.00
0.95
0.10
0.89
0.02
0.98
0.08
0.46
AtomVLA (Ours)
0.81
0.02
0.98
0.11
0.95
0.16
0.95
0.01
0.55
0.01
0.95
0.09
0.93
0.10
0.99
0.00
0.48
TABLE IV: Comparisons with state-of-the-art methods on the LIBERO-PRO benchmark [ 32 ] . Bold denotes the best performance among all methods.
Fig. 4: Visualization of real-world tasks. The top two rows illustrate basic tasks to stack bowls, place fruit into a basket, hang the cup, and open the drawer. The bottom two rows demonstrate hard, long-horizon tasks to fold a T-shirt and a towel.
Reward
Spatial
Object
Goal
Long
Avg.
SFT
94.1%
95.4%
92.4%
90.0%
93.0%
Post-Training
Esub(k)+D(k)
96.3%
98.0%
96.2%
93.5%
96.0%
Egoal(k)+D(k)
97.1%
97.9%
96.0%
93.2%
96.1%
Esub(k)+Egoal(k)+D(k)
96.4%
99.6%
97.6%
94.4%
97.0%
TABLE V: Ablation of reward formulations on LIBERO Benchmark [ 31 ] . Bold denotes the best performance among all methods.
TABLE VI: Experimental results of AtomVLA and baseline on six real-world Galaxea R1 Lite robot manipulation tasks. Bold denotes the best performance among all methods.
Fig. 6: The four types of variation introduced under the GE setting to evaluate the robustness of AtomVLA. (a) Object position variation. (b) Unseen distractor object. (c) Target height variation. (d) Instruction language variation.
Appendix figures & tables5 assets
Supplementary material from the paper’s appendix.
Appendix
Fig. 7: Atomic subtasks on the LIBERO Dataset.
Hyperparameter
Value
# GPUs
4× NVIDIA H100 (80GB VRAM)
learning rate (LR)
5e−4
batch size
8 (per GPU)
# epoch number
2 epoch
input images
1 third-person camera image, 1 wrist-mounted camera image
Recent advances in Visual-Language-Action (VLA) models have shown promising potential for robotic manipulation tasks. However, real-world robotic tasks often involve long-horizon, multi-step problem-solving and require generalization for continual skill acquisition, extending beyond single actions or skills. These challenges present significant barriers for existing VLA models, which use monolithic action decoders trained on aggregated data, resulting in poor scalability. To address these challenges, we propose AtomicVLA, a unified planning-and-execution framework that jointly generates task-level plans, atomic skill abstractions, and fine-grained actions. AtomicVLA constructs a scalable atomic skill library through a Skill-Guided Mixture-of-Experts (SG-MoE), where each expert specializes in mastering generic yet precise atomic skills. Furthermore, we introduce a flexible routing encoder that automatically assigns dedicated atomic experts to new skills, enabling continual learning. We validate our approach through extensive experiments. In simulation, AtomicVLA outperforms π0 by 2.4% on LIBERO, 10% on LIBERO-LONG, and outperforms π0 and π0.5 by 0.22 and 0.25 in average task length on CALVIN. Additionally, our AtomicVLA consistently surpasses baselines by 18.3% and 21% in real-world long-horizon tasks and continual learning. These results highlight the effectiveness of atomic skill abstraction and dynamic expert composition for long-horizon and lifelong robotic tasks. The project page is here.
Likui Zhang, Tao Tang, Zhihao Zhan +9
Sun Yat-sen University · Yinwang Intelligent Technology Co. Ltd. · Peng Cheng Laboratory
Vision-language-action (VLA) models are effective robot action executors, but they remain limited on long-horizon tasks due to the dual burden of extended closed-loop planning and diverse physical operations. We therefore propose VLAs-as-Tools, a strategy that distributes this burden across a high-level vision language model (VLM) agent for temporal reasoning and a family of specialized VLA tools for diverse local physical operations. The VLM handles scene analysis, global planning, and recovery, while each VLA tool executes a bounded subtask. To tightly couple agent planning with VLA tool execution in long-horizon tasks, we introduce a VLA tool-family interface that exposes explicit tool selection and in-execution progress feedback, enabling efficient event-triggered agent replanning without continuous agent polling. To obtain diverse specialized VLA tools that faithfully follow agent invocations, we further propose Tool-Aligned Post-Training (TAPT), which constructs invocation-aligned training units for instruction following and adopts tool-family residual adapters for efficient tool specialization. Across five simulated benchmarks, VLAs-as-Tools improves the success rate of pi05 by 8.9 percentage points on average and enhances invocation fidelity by 15.0 points in Non-biased Rate. On three real-world Franka tasks, it improves success rate from 16.7% to 58.3% under matched data and compute. Code will be released.
Zixing Lei, Changxing Liu, Yichen Xiong +5
Shanghai Jiao Tong University · Zhongguancun Academy · Beihang University
General-purpose robots must master long-horizon manipulation, defined as tasks involving multiple kinematic structure changes (e.g., attaching or detaching objects) in unstructured environments. While Vision-Language-Action (VLA) models offer the potential to master diverse atomic skills, they struggle with the combinatorial complexity of sequencing them and are prone to cascading failures due to environmental sensitivity. To address these challenges, we propose LiLo-VLA (Linked Local VLA), a modular framework capable of zero-shot generalization to novel long-horizon tasks without ever being trained on them. Our approach decouples transport from interaction: a Reaching Module handles global motion, while an Interaction Module employs an object-centric VLA to process isolated objects of interest, ensuring robustness against irrelevant visual features and invariance to spatial configurations. Crucially, this modularity facilitates robust failure recovery through dynamic replanning and skill reuse, effectively mitigating the cascading errors common in end-to-end approaches. We introduce a 21-task simulation benchmark consisting of two challenging suites: LIBERO-Long++ and Ultra-Long. In these simulations, LiLo-VLA achieves a 69% average success rate, outperforming Pi0.5 by 41% and OpenVLA-OFT by 67%. Furthermore, real-world evaluations across 8 long-horizon tasks demonstrate an average success rate of 85%. Project page: https://yy-gx.github.io/LiLo-VLA/.
Yue Yang, Shuo Cheng, Yu Fang +4
University of North Carolina at Chapel Hill · Georgia Institute of Technology · Carnegie Mellon University