FastOPD: On-Policy Distillation for Lightweight VLA Deployment

Yoojin Oh; Jeongsol Kim; Yeonwoo Seo; Jangho Park; Seonghyun Jin; Sunwoo Park; Youngmin Kim; Youngjun Jun

cs.ROOct 2, 2026

FastOPD: On-Policy Distillation for Lightweight VLA Deployment

Authors: Yoojin Oh, Jeongsol Kim, Yeonwoo Seo, Jangho Park, Seonghyun Jin, Sunwoo Park, Youngmin Kim, Youngjun Jun, +2 more

Organizations: KAIST · Sungkyunkwan University

Abstract

Vision-Language-Action (VLA) foundation models have scaled rapidly to enhance manipulation performance and generalizability, but this scaling incurs high computational costs that render real-world deployment increasingly challenging. Existing approaches typically mitigate this issue by designing smaller architectures or reducing the iterative denoising steps in flow-based policies. In this work, we propose FastOPD, a foundation-to-lightweight VLA framework that enables the practical deployment of large-scale VLAs through efficient on-policy distillation. Specifically, FastOPD adapts a flow map for single-state teacher supervision and combines it with a self-consistency objective to construct a compact student that learns the teacher dynamics. Furthermore, we theoretically demonstrate that minimizing this objective allows the distilled student to recover a distribution on par with that induced by an ideal few-step teacher model. We evaluate FastOPD across diverse foundation policies in simulation and real-world experiments. On LIBERO, FastOPD retains 84% of the performance of $π_{0.5}$ with only two inference steps, reducing inference latency by 78.1% while outperforming existing few-step distillation baselines in average success rate. With LingBot-VLA as the teacher, FastOPD improves the single-step success rate over the base student by 15.9 percentage points on RoboTwin 2.0. We further demonstrate its applicability to a World Action Model (WAM) and deploy a compact student distilled from MolmoAct2 on a real robot.

Figures & tables

Figure 1: Foundation-to-lightweight VLA policy distillation with FastOPD. (Left) Our method enables a small VLA to inherit large-scale VLA knowledge. (Right) FastOPD reaches comparable performance to conventional on-policy distillation ( Li et al., 2026a ) 5.7

\times

faster.

Panel of Figure 2 — Figure 2: Comparison between DiffusionOPD and FastOPD. (a) DiffusionOPD evaluates teacher at every denoising step. (b) FastOPD evaluates teacher at a single sampled state reached by a flow-map jump from noise. Blue arrows denote student transitions; red arrows denote teacher velocities.

Figure 3: Overview of FastOPD. (a) Flow Map Architecture: a frozen vision-language backbone with a lightweight action expert. A time projection layer $\psi(\cdot)$ fuses $(s,t)$ to condition the action expert on the two-time velocity ${\bm{u}}_{\theta}({\bm{x}}_{s},s,t)$ . (b) On-Policy Flow Map Distillation Loss: at a single on-policy jump $\tilde{{\bm{x}}}_{t}={\bm{f}}_{\theta}({\bm{x}}_{1},1,t)$ , the student’s diagonal velocity is matched to the teacher’s. (c) Self-Consistency Loss: a direct transition ${\bm{u}}_{\theta}({\bm{x}}_{s},s,t)$ is trained to agree with two steps through the midpoint $m$ .

Table 3: End-to-end inference latency (ms) on LIBERO and RoboTwin 2.0.

Figure 4: Success rate (%) on LIBERO during training: (a) FastOPD at different sampling steps, (b) comparison with baselines at 1 step.

Panel of Figure 5 — Figure 5: Real-world results on the pnp-plate task. (a) Success rate (%). (b) Task completion time (s), averaged over successful episodes.

Table 7

Figure 7: Ablation on action execution steps on LIBERO (NFE

=

2 for Ours and NFE

=

10 for

\pi_{0.5}

Appendix figures & tables13 assets

Supplementary material from the paper’s appendix.

Appendix

Benchmark	Model	Checkpoint
LIBERO	$\pi_{0.5}$	lerobot/pi05_libero_finetuned_v044
	SmolVLA	HuggingFaceVLA/smolvla_libero
	SmolVLA (OPD Init)	lerobot/smolvla_base
RoboTwin 2.0	LingBot-VLA	robbyant/lingbot-vla-4b-posttrain-robotwin
	Fast-WAM	lerobot/fastwam_robotwin_uncond_3cam_384
	SmolVLA	lerobot/smolvla_robotwin

Appendix

Table 7: Model checkpoints used in the experiments.

	LIBERO	RoboTwin 2.0		Real-world
Model
Teacher (Params)	$\pi_{0.5}$ (3.3B)	LingBot-VLA (4.2B)	Fast-WAM (6B)	MolmoAct2 (5B)
Student (Params)	SmolVLA + $\psi$ (451M)	SmolVLA + $\psi$ (451M)	SmolVLA + $\psi$ (451M)	SmolVLA + $\psi$ (451M)
Optimization
Optimizer	AdamW	AdamW	AdamW	AdamW
Adam $(\beta_{1},\beta_{2})$	$(0.9,0.95)$	$(0.9,0.95)$	$(0.9,0.95)$	$(0.9,0.95)$

Appendix

Table 8: Training and evaluation configurations for FastOPD.

Experiment	Hardware
LIBERO (main)	$1\times$ NVIDIA RTX 4090 (24 GB)
LIBERO (ablations)	$1\times$ NVIDIA RTX 4090 or RTX 3090 (24 GB each)
RoboTwin 2.0 (LingBot-VLA)	$2\times$ NVIDIA RTX 4090 (24 GB each)
RoboTwin 2.0 (Fast-WAM)	$1\times$ NVIDIA H200 (141 GB)
Real-world	$1\times$ NVIDIA B200 (180 GB)

Appendix

Table 9: Hardware used for training.

Figure 8: Robot setting. (a) Teleoperation setup for data collection, with lead arms, the follower YAM arm, and top-view and wrist cameras. (b) Close-up view of the evaluation setup. (c) Object-position initialization: the current top-camera frame is compared against a reference frame, and their pixel-wise difference is used to place objects at the same initial positions across trials.

Task	Language instruction	Episodes	Frames	Avg. Length (s)
pnp-plate	“Pick up the ball and place it on the plate.”	106	47,002	14.8
pnp-box	“Pick up the basketball and place it in the box.”	108	47,616	14.7
open-close	“Open the box, put the ball inside, and close it.”	53	44,780	28.2
open-close	“Open the box, put the basketball inside, and close it.”	55	46,834	28.4
open-place	“Open the box, take the ball out, and place it in the pot.”	52	41,637	26.7
open-place	“Open the box, take the ball out, and place it on the tray.”	50	40,676	27.1

Appendix

Table 10: Real-world dataset statistics. Average episode length is computed at 30 fps.

Figure 9: Real-world manipulation task setups on the YAM robot arm. Left two columns show the initial scene and the right show the robot during execution.

	MolmoAct2	SmolVLA (warm-up)	Ours
Execution	async (re-plan every 0.5 s)	sync	sync
Sampler	Euler, 10 steps	Euler, 4 / 10 steps	flow map, 4 steps
Actions executed per chunk	$\sim$ 15	30	30
Chunk blending	linear (2–10 steps)	none	none
Inference latency	300–450 ms	118 / 178 ms	120 ms
Re-planning rate	1.8–1.9 Hz	0.89 / 0.86 Hz	0.87 Hz

Appendix

Table 11: Real-world policy execution settings. All policies predict 30-step action chunks (1.0 s) and send commands at 30 Hz. For SmolVLA, values are reported for 4 and 10 denoising steps. Inference latency is measured on the robot side, including HTTP communication.

Method	NFE	Spatial	Object	Goal	Long	Avg.
OPD init.	1	78.0	94.2	78.2	40.2	72.7
	2	84.4	95.4	87.8	45.0	78.2
	4	58.0	59.8	53.8	5.0	44.2
	10	82.6	94.6	89.0	48.2	78.6
+ FastOPD	1	84.2	92.8	90.8	54.0	80.5
	2	84.6	95.2	94.2	57.8	83.0

Appendix

Table 12: Per-suite results of OPD-based initialization. Success rates (%) on LIBERO.

Figure 10: Success rate (%) on LIBERO over training iterations with OPD-based initialization.

Task	LingBot-VLA		Fast-WAM		SmolVLA		Ours (LingBot-VLA)		Ours (Fast-WAM)
Task	Clean	Random	Clean	Random	Clean	Random	Clean	Random	Clean	Random
adjust_bottle	98	90	100	100	100	98	100	98	100	98
beat_block_hammer	78	78	100	96	66	66	56	40	76	78
blocks_ranking_rgb	96	92	100	100	50	46	50	58	30	30
blocks_ranking_size	74	66	88	96	20	22	28	26	20	10
click_alarmclock	70	42	100	100	90	94	34	50	92	96

Appendix

Table 13: Per-task success rates (%) on RoboTwin 2.0 at 10 denoising steps, under clean and randomized settings. Ours (X) denotes our method trained with X as the teacher.

Task	Ours (LingBot-VLA)						Ours (Fast-WAM)
	1-step		2-step		4-step		1-step		2-step		4-step
	Clean	Random	Clean	Random	Clean	Random	Clean	Random	Clean	Random	Clean	Random
adjust_bottle	94	96	100	98	100	98	94	98	100	96	100	98
beat_block_hammer	72	58	84	68	66	66	56	50	76	68	80	74
blocks_ranking_rgb	34	32	38	44	52	56	34	32	44	32	32	32
blocks_ranking_size	6	6	20	18	28	16	12	8	12	10	20	14

Appendix

Table 14: Per-task success rates (%) of our method with LingBot-VLA and Fast-WAM as teachers on RoboTwin 2.0 at 1, 2, and 4 denoising steps, under clean and randomized settings. Ours (X) denotes our method trained with X as the teacher.

Figure 11: Qualitative results on RoboTwin 2.0. FastOPD distilled from LingBot-VLA (NFE=1) executing four tasks under the clean (top) and randomized (bottom) settings, where the randomized setting introduces cluttered distractors, textured tabletops, and varied backgrounds.

Figure 12: Qualitative results on the real-world task. Rollouts of the MolmoAct2 teacher and the student after Stage 1 (initialization) and Stage 2 (FastOPD) on the YAM robot arm, in which the robot picks up a ball and places it on a plate. Time progresses from left to right; insets show the wrist-camera view.

Explore similar work

Offline Semantic Guidance for Efficient Vision-Language-Action Policy Distillation
May 15, 2026Jin Shi, Brady Zhang, Yishun LuVLM Distillation Efficient VLA Model Inference
Department of Mechanical Engineering University College London London, United Kingdom · Department of Engineering Science University of Oxford Oxford, United Kingdom
VLA-OPD: Bridging Offline SFT and Online RL for Vision-Language-Action Models via On-Policy Distillation
Mar 27, 2026Zhide Zhong, Haodong Yan, Junfeng Li +3Vision-Language-Action Models On-Policy Distillation
HKUST (GZ)
Reducing Temporal Redundancy for Efficient Vision-Language-Action Inference
Jul 14, 2026Yuzhou Wu, Yuxin Zheng, Muchun Niu +6Efficient VLA Model Inference Efficient VLA Models
1Tianji KernalMind co ltd · 2Tsinghua Shenzhen International Graduate School, Tsinghua University, Shenzhen, China · 3Shanghai Jiao Tong University, Shanghai, China +2

May 15, 2026J/K move · Enter open · S save

Offline Semantic Guidance for Efficient Vision-Language-Action Policy Distillation

Jin Shi, Brady Zhang, Yishun Lu

Department of Mechanical Engineering University College London London, United Kingdom · Department of Engineering Science University of Oxford Oxford, United Kingdom

VLM Distillation Efficient VLA Model Inference Visuomotor Policy Learning Efficient VLA Models Robotic Manipulation Vision-Language-Action Models+1

Open paper AskPDF arXiv

Billion-parameter Vision-Language-Action (VLA) policies have recently shown impressive performance in robotic manipulation, yet their size and inference cost remain major obstacles for real-time closed-loop control. We introduce VLA-AD, a distillation framework that uses a Vision-Language Model as an offline semantic supervisor to transfer large VLA teachers into lightweight student policies. Instead of relying only on low-level action imitation, VLA-AD augments teacher-provided 7-DoF action targets with high-level semantic guidance, including task phase anchors and multi-frame operating-direction descriptions. These auxiliary signals are used only during training: at test time, the student policy runs independently, with neither the VLA teacher nor the VLM required. We evaluate VLA-AD on three LIBERO benchmark suites. Using OpenVLA-7B as the teacher, our method produces a 158M-parameter student, yielding a

44\times

reduction in model size while matching the teacher with only a

0.27\%

average relative gap. The resulting policy runs at 12.5 Hz on an RTX 4090, achieving a

3.28\times

inference speedup over OpenVLA-7B. We further show that the same semantic distillation pipeline generalizes to a different

π_{0.5}

-4B teacher, where the student outperforms the teacher on two suites and remains within

0.53\%

on libero\_goal. Additional analysis indicates that phase-level supervision and multi-frame directional cues make the student less sensitive to noisy teacher actions, such as erroneous high-frequency gripper changes. Overall, VLA-AD demonstrates that offline semantic guidance from VLMs can substantially improve the efficiency, robustness, and deployability of VLA policy distillation.

Abstract

Explore similar work

Offline Semantic Guidance for Efficient Vision-Language-Action Policy Distillation

VLA-OPD: Bridging Offline SFT and Online RL for Vision-Language-Action Models via On-Policy Distillation

Reducing Temporal Redundancy for Efficient Vision-Language-Action Inference