V-IRL is an open-source platform that bridges the gap between digital and physical worlds, enabling AI agents to interact with the real world in a virtual yet realistic environment. The platform uses real geospatial data and street-view imagery to provide agents with rich sensory grounding and perception. It serves as a flexible playground for developing agents that can perform various practical tasks and as a vast testbed for measuring progress in capabilities spanning perception, decision-making, and interaction with real-world data across the globe. V-IRL allows agents to navigate real-world locations, access up-to-date information about their surroundings, and perform practical tasks. It integrates with arbitrary geospatial platforms and APIs, and provides a vast sea of visual data for researchers to evaluate vision models on realistic data distributions. The platform has been used to develop diverse exemplar agents, each solving a unique and practical task. These agents demonstrate the platform's versatility and adaptability. The paper also presents three V-IRL benchmarks: two evaluating vision-language models on open-world vision tasks and one evaluating end-to-end agent performance. The benchmarks are designed to test core component models and agent capabilities using scalable, truly open-world data. The results show that open-world detectors like GroundingDINO, Owl-ViT, and GLIP are biased towards certain place types, while CLIP (with GLIP proposal) can identify a broader spectrum of place types. The results also show that vision models perform better on place VQA over place-type recognition, suggesting direct prompts about human intention could be more effective for intention-driven tasks. The platform also demonstrates the effectiveness of V-IRL in various real-world scenarios, including urban assistance, language-driven tasks, and collaborative tasks. The paper highlights the importance of V-IRL in advancing AI capabilities in perception, decision-making, and real-world data interaction. It also discusses the ethical and privacy concerns associated with the platform, emphasizing the need for responsible AI development. Overall, V-IRL provides a valuable tool for researchers to study bias and develop more effective AI agents.V-IRL is an open-source platform that bridges the gap between digital and physical worlds, enabling AI agents to interact with the real world in a virtual yet realistic environment. The platform uses real geospatial data and street-view imagery to provide agents with rich sensory grounding and perception. It serves as a flexible playground for developing agents that can perform various practical tasks and as a vast testbed for measuring progress in capabilities spanning perception, decision-making, and interaction with real-world data across the globe. V-IRL allows agents to navigate real-world locations, access up-to-date information about their surroundings, and perform practical tasks. It integrates with arbitrary geospatial platforms and APIs, and provides a vast sea of visual data for researchers to evaluate vision models on realistic data distributions. The platform has been used to develop diverse exemplar agents, each solving a unique and practical task. These agents demonstrate the platform's versatility and adaptability. The paper also presents three V-IRL benchmarks: two evaluating vision-language models on open-world vision tasks and one evaluating end-to-end agent performance. The benchmarks are designed to test core component models and agent capabilities using scalable, truly open-world data. The results show that open-world detectors like GroundingDINO, Owl-ViT, and GLIP are biased towards certain place types, while CLIP (with GLIP proposal) can identify a broader spectrum of place types. The results also show that vision models perform better on place VQA over place-type recognition, suggesting direct prompts about human intention could be more effective for intention-driven tasks. The platform also demonstrates the effectiveness of V-IRL in various real-world scenarios, including urban assistance, language-driven tasks, and collaborative tasks. The paper highlights the importance of V-IRL in advancing AI capabilities in perception, decision-making, and real-world data interaction. It also discusses the ethical and privacy concerns associated with the platform, emphasizing the need for responsible AI development. Overall, V-IRL provides a valuable tool for researchers to study bias and develop more effective AI agents.