Congratulations to the SingaX team on clinching second place in the Embodied Agent Interface (EAI) Challenge held at the Neural Information Processing Systems (NeurIPS) 2025 conference in San Diego, California.

The EAI Challenge attracted teams from around the world to compete in developing embodied, agentic systems capable of interpreting natural language instructions and executing complex tasks in simulated environments. Participants were evaluated across four demanding modules: Goal Interpretation, Subgoal Decomposition, Action Sequencing, and Transition Modelling.

The challenge was particularly rigorous, as embodied agents are required to reason over long-horizon instructions, track intermediate states, and generate action sequences that are both semantically grounded and executable. Traditional approaches often fall short due to brittle prompt structures and limited awareness of system-level constraints, resulting in parsing errors or invalid action sequences.

Despite these challenges, the SingaX team introduced an innovative iterative prompt induction framework that automatically learns from historical experience. By analysing failure patterns from development tasks, the framework progressively refines task-specific instructions, enabling the agent to avoid prior logical inconsistencies in unseen problems. This training-free and cost-efficient approach proved to be both highly effective and generalisable.

With an impressive average score of 84.32, the SingaX team ranked second out of 48 teams on the leaderboard. Their work demonstrates a scalable approach to embodied task planning and offers valuable insights into integrating large language models within interactive agent pipelines.

The SingaX team comprises researchers from the Agency for Science, Technology and Research (A*STAR), National University of Singapore (NUS) and Nanyang Technological University, Singapore (NTU Singapore).

Team Members:

• A*STAR Computing and Information Science (ACIS) Scholars: Niu Xinyuan (NUS), Chen Zhiliang (NUS), Vernon Toh (NTU Singapore) and Li Yanchao (NTU Singapore)
• Supervisors: Nancy F. Chen (A*STAR) and Liu Zhengyuan (A*STAR)
  

More on the EAI Challenge.