Vision-Language-Action (VLA) models such as π0 have demonstrated remarkable generalization across diverse fixed-base manipulators. However, transferring these foundation models to aerial platforms remains an open challenge due to the fundamental mismatch between the quasi-static dynamics of fixed-base arms and the underactuated, highly dynamic nature of flight. In this work, we introduce AirVLA, a system that investigates the transferability of manipulation-pretrained VLAs to aerial pick-and-place tasks. We find that while visual representations transfer effectively, the specific control dynamics required for flight do not. To bridge this "dynamics gap" without retraining the foundation model, we introduce a Payload-Aware Guidance mechanism that injects payload constraints directly into the policy's flow-matching sampling process. To overcome data scarcity, we further utilize a Gaussian Splatting pipeline to synthesize navigation training data. We evaluate our method through a cumulative 460 real-world experiments which demonstrate that this synthetic data is a key enabler of performance, unlocking 100% success in navigation tasks where directly fine-tuning on teleoperation data alone attains 81% success. Our inference-time intervention, Payload-Aware Guidance, increases real-world pick-and-place task success from 23% to 50%. Finally, we evaluate the model on a long-horizon compositional task, achieving a 62% overall success rate. These results suggest that pre-trained manipulation VLAs, with appropriate data augmentation and physics-informed guidance, can transfer to aerial manipulation and navigation, as well as the composition of these tasks.
Figures & tables
Fig. 1: Custom gripper installed on the Starling 2 Max drone. The design prioritizes low weight for extended flight time while maintaining sufficient grip strength for the target objects.
Fig. 2: Examples of synthetic trajectories used to generate training data. Nominal rollouts and corrective rollouts overlay sampled trajectories in the gate-manipulation scene.
Method
Penguin Grasp
Navigation (Non-Synthetic)
Navigation (Synthetic)
Pick
Place
Gate
Hover
Gate
Hover
π0 naive
50.0
0.0
50.0
60.0
45.0
100.0
π0 + RTC
85.0
23.5
80.0
81.2
95.0
100.0
π0 + RTC w/ payload-aware guidance (ours)
100.0
50.0
–
–
–
–
ACT
0.0
0.0
0.0
0.0
0.0
0.0
Diffusion Policy
10.0
0.0
15.0
0.0
0.0
0.0
TABLE I: Single Task Performance Benchmarks (%). Each method was evaluated across twenty trials per task.
Data Setting
Method
Gate
Hover
Pick
Place
No Synthetic
π0 naive
35.0
85.7
42.9
0.0
π0 + RTC
80.0
100.0
81.2
15.4
π0 + RTC w/ payload-aware guidance (ours)
70.0
100.0
100.0
35.7
ACT
0.0
0.0
0.0
0.0
Diffusion Policy
5.0
0.0
0.0
0.0
Synthetic
π0 naive
70.0
85.7
25.0
0.0
TABLE II: Compositional Navigate-then-Grasp Success (%). Each method was evaluated across twenty trials per task.
Grasp (Object Variation)
Navigation (Gate Locations)
Object
Pick
Place
Location
Gate
Hover
Chips
10.0
0.0
Front
0.0
–
Sandwich
70.0
57.1
Left
0.0
–
Box
30.0
33.3
Right
40.0
100.0
TABLE III: Out-of-Distribution (OOD) Robustness (%). Success rates for variations in object class and gate positioning. Place and Hover are reported as conditional percentages.
Fig. 3: Out-of-distribution objects used for generalization probes. From left to right: goomba, box, chips, and sandwich. The policy was fine-tuned on a penguin but evaluated on these novel objects to test object-level generalization beyond training instances.
Fig. 4: Trajectory visualization for the left gate task. The top three rows display real-world RGB observations from external, forward, and downward cameras. The bottom two rows show corresponding synthetic views rendered from Gaussian Splats.
Aerial manipulators extend robotic manipulation into 3D workspaces that are difficult for ground-based robots to access, creating new opportunities for general-purpose manipulation. However, extending Vision-Language-Action (VLA) models to aerial robots introduces distinct challenges due to the tight coupling between manipulation and flight, continuously changing observations, and safety-critical physical interactions. These challenges demand diverse training data and systematic policy evaluation, yet collecting demonstrations and evaluating policies directly on physical aerial platforms are costly, difficult to scale, and hard to repeat under controlled conditions. We present AeroManip-VLA, a scalable benchmark for aerial VLA data generation and policy evaluation. AeroManip-VLA provides a GPU-accelerated simulation framework with low-level payload-aware flight and manipulation control in massively parallel environments. Building on this framework, we combine reusable reinforcement learning policies with expert task rules to automatically generate demonstrations without human teleoperation across diverse objects, environments, and randomized initial conditions. The generated data include basic skills such as grasping and placing, as well as long-horizon tasks that require both navigation and manipulation. We further introduce automated event labeling and trajectory categorization to filter demonstrations. These mechanisms enable fine-grained analysis of task progress, behavioral outcomes, and safety-related failures. Finally, we evaluate a range of imitation learning and VLA baselines across different task settings, revealing their performance characteristics and failure modes. Together, AeroManip-VLA enables scalable aerial manipulation data generation, structured trajectory analysis, and systematic VLA evaluation in simulation prior to real-world deployment.
Rui Huang, Yanlin Mu, Lidong Li +3
National University of Singapore · Beijing Institute of Technology
Vision-language-action (VLA) models enable task-conditioned interaction, but extending them to scene-scale aerial manipulation remains challenging due to costly whole-body demonstrations, latency-induced action-state misalignment, and cross-site behavior composition. We present a unified framework for synthetic policy training and scene-scale execution on articulated uncrewed aerial manipulators (UAMs). A scene-reconfigurable pipeline synthesizes task-conditioned, kinodynamically feasible trajectories and synchronized multiview observations for VLA training without physical-platform demonstrations. Measured-progress-aligned realization (MPAR) aligns asynchronously returned action chunks with measured execution progress and realizes them as continuous, dynamically feasible trajectories. A relational Scene Graph grounds language goals to object instances and feasible interaction regions, while topology-guided transfer connects local behaviors across sites. Local VLA skills achieve 39/60 successes (65.0%) in simulation under oracle target and feasible-handoff conditions. Under 500-ms added latency, with and without a transient command-update stall, MPAR reduces median takeover phase error by 0.212 s over nominal-time alignment. The complete system completes 21/50 simulated multi-site missions (42.0%) and is further validated on a physical articulated UAM.
Weixiang Guo, Rui Jin, Haotian Jin +7
School of Electrical and Electronic Engineering, Nanyang Technological University, Singapore 639798 · College of Electronics and Information Engineering, Shanghai Institute of Intelligent Science and Technology, Tongji University, Shanghai 201804, China · School of Aeronautic Science and Engineering, Beihang University, Beijing 100191, China +1
Vision-Language-Action (VLA) models offer a promising end-to-end paradigm for unmanned aerial vehicles (UAVs) to accomplish complex tasks specified by fine-grained instructions. However, standard supervised fine-tuning (SFT) suffers from data scarcity, limited generalization, and weak supervision for nuanced and complicated human intents. Reinforcement fine-tuning offers a natural way to mitigate these challenges and align policy behaviors with human intents through designable feedback, but applying it to aerial navigation remains challenging due to inefficient exploration in expansive continuous spaces. To address these challenges, we introduce an efficient reinforcement learning (RL) framework for VLA-based aerial navigation. At its core, we propose EG-GRPO (Expert-Guided Group Relative Policy Optimization) to augment online rollouts with few-shot expert data. Additionally, we design a heterogeneous pipeline enabling parallel simulation and inference, which reduces rollout time by 43.5%. Across multiple tasks specified by complex human intents, EG-GRPO improves the success rate to 2.13x that of the SFT baseline, while improving intent alignment performance by 60.9%. These results demonstrate that our framework can move aerial navigation toward precise intent-aligned flight.