AGIBOT Releases Fourth AGIBOT WORLD 2026 Dataset Focused on Robot Decision-Making

AGIBOT has announced the open-source release of its Physical Reasoning dataset, the fourth release of AGIBOT WORLD 2026, shifting task design focus from manipulation difficulty to decision-making complexity.

AGIBOT WORLD 2026 is described as the first open-source dataset to cover the full range of embodied AI research, organized around five research themes with dedicated data-collection methods and a fine-grained annotation system. Release 4 captures decision-making and execution processes for the same goal across different states and constraints.

For example, when asked to pack for going out the next day, an agent first checks the weather forecast using a tool, then determines which items are needed — sunscreen for sun, an umbrella for rain — and directs a robot to pick and place them. The dataset covers over 10 categories of logical reasoning scenarios and deliberately includes multiple real-world states and execution branches for the same task.

The dataset addresses nine core reasoning abilities: attribute-constrained selection, context-conditioned selection, manipulation feasibility assessment, action precondition reasoning, goal-directed search, state and quantity reasoning, long-horizon planning, memory and state maintenance, and subtask ordering reasoning. Tasks are organized into Easy, Medium and Hard difficulty levels, with execution outcomes annotated and successful and erroneous samples given equal weight.

At the annotation level, high-level instructions describe the complete goal, while task frames, subtask segments and action steps capture task structure. Dense Prompts specify the target object, manipulation region, bimanual role assignment, motion process, gripper state and expected outcome, aligned with real-robot actions. Spatial annotations including bounding boxes, points and trajectory lines support visual-language grounding.

The dataset has already been used to support AGIBOT's upcoming agent framework and to train its next-generation foundation model. AGIBOT's "Three Intelligences in One" architecture integrates locomotion, interaction and manipulation intelligence, and in September 2026 the company announced its 20,000th robot had rolled off the production line.