Jacqueline Staub

Chao Wen, Jacqueline Staub, Adish Singla

Vision-language models (VLMs) have been explored for visual programming, where they generate code to solve visual tasks. However, most prior work focuses on visual programming for productivity; it remains unclear how well current VLMs perform on education-oriented visual programming and what factors limit their performance. To bridge this gap, we introduce TurtleAI, a benchmark containing 823 tasks curated based on real-world visual programming tasks in the Turtle Graphics domain. Solving these tasks requires models to perceive geometric patterns, reason about spatial relationships, and synthesize Python code that faithfully reproduces geometric patterns. We evaluate 20+ VLMs, including GPT-5, GPT-4o, and Qwen2-VL-72B, and find that they struggle significantly, with most achieving success rates below 30%. To address these limitations, we propose a data generation technique that requires only a small set of seed samples. Fine-tuning Qwen2-VL-72B on the resulting synthetic data yields an improvement of about 20% on real-world tasks. Our failure analysis reveals that GPT-4o struggles with spatial reasoning and precise visual replication, whereas fine-tuning primarily improves the alignment between visual reasoning and code implementation.

Chao Wen, Jacqueline Staub, Adish Singla

Large language and multimodal models have shown remarkable success on various benchmarks focused on specific skills such as general-purpose programming, math word problem-solving, and visual question answering. However, it is unclear how well these models perform on tasks that require a combination of these skills. In this paper, we curate a novel program synthesis benchmark based on the real-world tasks in the XLogoOnline visual programming environment. Each task requires a combination of different skills such as spatial planning, basic programming, and logical reasoning. Our evaluation shows that current state-of-the-art models like GPT-4V and Llama3-70B struggle to solve these tasks, achieving only 20% and 2.35% success rates, respectively. Next, we develop a fine-tuning pipeline to boost the performance of models by leveraging a large-scale synthetic training dataset with over 80,000 tasks. Moreover, we showcase how emulator-driven feedback can be used to design a curriculum over training data distribution, through which a fine-tuned Llama3-8B drastically outperforms GPT-4V and Llama3-70B models. Finally, we provide an in-depth failure analysis to understand the limitations of different models. We will publicly release the benchmark for future research on program synthesis in visual programming.