Leaderboard

https://airtable.com/appeN9J5OcPmfLzgf/shruaokvPDslOxufp

Challenge Tasks Overview

For additional details on each of the tasks and their corresponding datasets, see the Competition Wiki.

Automatic Speech Recognition (ASR): Teams will train an ASR model on our provided dataset. For Novice participants, the entire dataset is in English. For Advanced participants, the dataset is in English, Malay, Tamil, and Chinese (Mandarin). Note that the underlying transcripts are all set in the same world as our NLP dataset, and thus contain many original slang terms and phrases.

Computer Vision (CV) and Adversarial Noising: Teams will train an object detection model on our provided dataset. Additionally, they will also develop an “Adversarial Noising” pipeline, which will add noise of various types to the images that opponents’ models will receive. Note that the added noise must remain within specific thresholds, as determined by a noise thresholding script given to participants. If any of these thresholds are exceeded, the original un-noised image will be given to your opponents instead.

Natural Language Processing (NLP): Teams will develop a Retrieval-Augmented Generation (RAG) system to answer questions derived from a large corpus of text documents. The exact documents, and the details contained within them, will change depending on each phase of the competition. The documents are all based in an original fictional cyberpunk world that exists nowhere else, so you must extract the information specifically contained within the documents and ensure your language model does not hallucinate.

Autonomous Exploration (AE): Teams will develop an agent to autonomously explore a gridworld environment. Agents can move around to collect points from reconnaissance and mission tasks. You can collect resources from dedicated tiles as well as passively over time, which allow you to deploy bombs to destroy the environment and to attack other players. Teams additionally have control and visibility from their base, which opponents can also attack for points.

Submission Process

For Qualifiers, all submissions will be handled through the GCP Online Development Environment.

Additional instructions on how to work within the development environment are available on the Competition Wiki.
For each of the qualifier competition tasks (i.e. ASR, CV, noising, NLP, and AE), teams will be expected to use CLI commands to submit a Docker image with your submitted model packaged inside, for a total of 5 distinct Docker images to be submitted. Detailed step-by-step instructions are included in the competition base code repository. These submitted Docker images will be automatically run and evaluated against our hidden test cases, with your scores automatically being updated to our live competition leaderboard.

<aside> <img src="https://cdn-icons-png.flaticon.com/512/7154/7154370.png" alt="https://cdn-icons-png.flaticon.com/512/7154/7154370.png" width="40px" /> REMINDER: All teams are to submit before 24 May 2026, 11:59:59 PM SGT/GMT +8 for the Virtual Qualifiers. If you face any issues or require help during submission, ask around on the #hackoverflow channel on the TIL-AI Discord Server, message us in your team channel on the TIL-AI Discord Server, or email us at [email protected].

</aside>

For Semi-Finals/Finals, all Docker image submissions will also be handled through the GCP Online Development Environment.

For each of the qualifier competition tasks (i.e. ASR, CV, noising, NLP, and AE) as well as the autonomy and main orchestration server, teams will be expected to use CLI commands to submit a Docker image with your submitted model packaged inside, for a total of 5 distinct Docker images to be submitted. Detailed step-by-step instructions are included in the competition base code repository.
Participants are also able to submit their models for evaluation like for the Qualifiers, but they will be informed of their scores privately, and new scores will not be reflected on the leaderboard.
Participants will then get hands-on with the robots being deployed at the Semi-Finals + Finals at an in-person testing session on June 10th 2026 at MBS; booking links for this will be released for this on 1 June 2026, 12:00 PM. Participants will be allowed to make additional changes to their submitted images during this session as well.