Robotic Photography System based on Vision-Language-Action Model

Song-Li Wu; Chubin Zhang; Wenbo Ding; Yansong Tang

cs.ROOct 19, 2026

Robotic Photography System based on Vision-Language-Action Model

Authors: Song-Li Wu, Chubin Zhang, Wenbo Ding, Yansong Tang

Organizations: Shenzhen International Graduate School, Tsinghua University

Abstract

Modern robotic photography requires understanding photographic intent, planning camera motions, and producing aesthetically pleasing images in complex environments. We present RoboPhoto-VLA, a unified Vision-Language-Action framework that jointly models perception, reasoning, and camera control for autonomous robotic photography. RoboPhoto-VLA incorporates PhotoCoT for structured action planning and Aesthetic Direct Preference Optimization (ADPO) to align camera behaviors with human aesthetic preferences. To facilitate training and evaluation, we construct OpenPhotography, a robotic photography dataset with language-guided demonstrations. Extensive real-world experiments demonstrate that RoboPhoto-VLA produces photographs better aligned with human aesthetic preferences while generalizing effectively to unseen subjects, viewpoints, poses, and complex photographic instructions.

Figures & tables

Fig. 1: RoboPhoto-VLA first receives the user’s language request alongside a target scene (upper left: the user asks to capture a photo of themselves eating an apple). The system then detects the user, identifies incomplete framing of the pose, and initiates adjustments to its position and angle (upper right). Next, RoboPhoto-VLA precisely adjusts its robotic arm and camera orientation to align with the desired composition (lower left). The lower-right panel displays the final unretouched photograph captured by RoboPhoto-VLA.

Fig. 2: RoboPhoto-VLA model architecture. Given an image observation and a language instruction, the model predicts executable robot camera actions. The architecture follows a two-stage training paradigm: (1) Supervised Fine-Tuning (SFT) with PhotoCoT for structured photographic grounding and multi-step action planning; (2) Reinforcement Learning (RL) with ADPO for refining action selection to better align with human aesthetic preferences.

Experiment	Subtask	RT-2-X			$\pi_{0}$			OpenVLA			ANSEL			Snapbot			RoboPhoto-VLA
Experiment	Subtask	QS	AS	IIA	QS	AS	IIA	QS	AS	IIA	QS	AS	IIA	QS	AS	IIA	QS	AS	IIA
Cross-subject	Single-Person	0.09	0.05	0.02	0.37	0.48	0.35	0.29	0.41	0.32	0.63	0.78	0.65	0.72	0.61	0.69	0.86	0.91	0.87
Cross-subject	Multi-Person	0.03	0.07	0.01	0.21	0.31	0.32	0.27	0.36	0.38	0.75	0.66	0.71	0.64	0.77	0.68	0.89	0.90	0.91
Cross-view	45°	0.03	0.04	0.06	0.22	0.46	0.30	0.21	0.23	0.37	0.62	0.73	0.70	0.67	0.76	0.65	0.73	0.85	0.89
	90°	0.07	0.08	0.09	0.35	0.49	0.36	0.20	0.45	0.39	0.74	0.63	0.78	0.66	0.71	0.69	0.88	0.81	0.86
	135°	0.00	0.03	0.03	0.24	0.40	0.31	0.22	0.31	0.30	0.64	0.75	0.62	0.73	0.70	0.67	0.85	0.83	0.85

TABLE I: Performance comparison of RoboPhoto-VLA with other models for photography tasks. ”HoP” denotes Human-only Photography and ”HOP” denotes Human–Object Photography

Fig. 3: Instruction: Photograph me looking serious.

Experiment	Subtask	OpenVLA			w/ SFT			w/ ADPO			w/ DPO			RoboPhoto-VLA
Experiment	Subtask	QS	AS	IIA	QS	AS	IIA	QS	AS	IIA	QS	AS	IIA	QS	AS	IIA
Cross-subject	Single-Person	0.29	0.41	0.32	0.53	0.57	0.47	0.73	0.81	0.74	0.67	0.61	0.65	0.76	0.84	0.79
Cross-subject	Multi-Person	0.27	0.36	0.38	0.38	0.42	0.56	0.74	0.72	0.75	0.68	0.52	0.63	0.81	0.83	0.84
Cross-view	45°	0.21	0.23	0.37	0.26	0.52	0.41	0.63	0.72	0.81	0.56	0.53	0.69	0.73	0.85	0.89
	90°	0.20	0.45	0.39	0.46	0.62	0.44	0.78	0.77	0.73	0.71	0.63	0.66	0.88	0.81	0.86
	135°	0.22	0.31	0.30	0.31	0.58	0.42	0.71	0.75	0.78	0.63	0.55	0.63	0.85	0.83	0.85

TABLE II: Ablation Study

Fig. 4: We evaluate the model under cross-subject and cross-view generalization settings. For cross-subject evaluation, we employ the instructions “Capture me smiling naturally” and “Photograph me laughing with a friend” to assess performance in single-person and multi-person scenarios, respectively. For cross-view evaluation, we use the instruction “Take a photo of me drinking water” to examine robustness to viewpoint variations.

Vision-Language-Action Models

Vision-language-action models turn visual observations and natural-language instructions into actions for robots and other embodied agents.

See all 1,190 papers

PhotoAgent: A Robotic Photographer with Spatial and Aesthetic Understanding
Mar 24, 2026Lirong Che, Zhenfeng Gan, Yanbo Chen +2Multimodal Agents Vision-Based Robot Control
Center for Artificial Intelligence and Robotics, Shenzhen International Graduate School, Tsinghua University, Shenzhen 518055, China
VersaCamVLA: Camera-Configurable VLA Policies for Robotic Manipulation
Oct 8, 2026Boyao Han, Chen Shi, Jingjing Qian +2Vision-Language-Action Models Robotic Manipulation
The Chinese University of Hong Kong, Shenzhen · Shenzhen Loop Area Institute · Harbin Institute of Technology, Shenzhen
PhotoFlow: Agentic 3D Virtual Photography Missions
May 22, 2026Jiarui Guo, Haojia Wei, Yiming Zhang +5Tool-Using Vision-Language Agents 3D Scene Understanding
Shanghai Jiao Tong University · Northeastern University · University of California, Los Angeles +3

Mar 24, 2026J/K move · Enter open · S save

PhotoAgent: A Robotic Photographer with Spatial and Aesthetic Understanding

Lirong Che, Zhenfeng Gan, Yanbo Chen, Junbo Tan+1

Center for Artificial Intelligence and Robotics, Shenzhen International Graduate School, Tsinghua University, Shenzhen 518055, China

Multimodal Agents Vision-Based Robot Control

Open paper AskPDF arXiv

Embodied agents for creative tasks like photography must bridge the semantic gap between high-level language commands and geometric control. We introduce PhotoAgent, an agent that achieves this by integrating Large Multimodal Models (LMMs) reasoning with a novel control paradigm. PhotoAgent first translates subjective aesthetic goals into solvable geometric constraints via LMM-driven, chain-of-thought (CoT) reasoning, allowing an analytical solver to compute a high-quality initial viewpoint. This initial pose is then iteratively refined through visual reflection within a photorealistic internal world model built with 3D Gaussian Splatting (3DGS). This ``mental simulation'' replaces costly and slow physical trial-and-error, enabling rapid convergence to aesthetically superior results. Evaluations confirm that PhotoAgent excels in spatial reasoning and achieves superior final image quality.

Abstract

Vision-Language-Action Models

Explore similar work

PhotoAgent: A Robotic Photographer with Spatial and Aesthetic Understanding

VersaCamVLA: Camera-Configurable VLA Policies for Robotic Manipulation

PhotoFlow: Agentic 3D Virtual Photography Missions