About

The IMU Odometry Challenge — run on Kaggle as the TartanIMU Challenge: Multi-Platform Inertial Odometry — is hosted by CMU AirLab, with organization led by the TartanIMU and Super Odometry contributors.

The challenge is the official benchmark of the IROS 2026 Workshop: “Beyond Exteroception: Interoceptive Perception for Resilient Robotics” (Sept 27 – Oct 1, 2026, Pittsburgh). Top entries will be invited to present their methods at the workshop. The competition does not award a cash prize — it’s scored on Kudos rather than points or medals.

Motivation

Learning-based inertial methods have shown strong progress, but almost always on a single platform, typically a pedestrian carrying a phone. Whether one inertial-perception method can generalize across embodiments with completely different dynamics, frequency content, and motion priors is still an open question.

The twist that defines this challenge: you must submit one unified model that handles all four platforms at once. The test set is anonymized — no platform label, no platform-revealing trajectory ids — so per-platform experts switched at inference time are not an option. This is a cross-embodiment generalization benchmark, not four separate regressions.

This challenge is designed to provide:

  • A common benchmark for cross-platform, single-model inertial odometry
  • Shared train/val resources (with platform labels) and an anonymized, platform-blind test evaluation
  • A transparent evaluation path for comparing new methods

Technical Context

This challenge is aligned with the Tartan IMU direction:

  • Large-scale multi-platform IMU pretraining
  • Efficient adaptation to unseen domains
  • Online adaptation for deployment-time robustness

The benchmark is intended to surface concrete modeling, adaptation, and evaluation questions that can feed into the workshop discussion around robot self-sensing and robustness beyond exteroceptive perception alone.

Organizing Team

Challenge organization is coordinated by CMU AirLab, with benchmark, dataset, and baseline development led by TartanIMU and Super Odometry contributors. The effort brings together researchers from Carnegie Mellon University and Amazon FAR.

Contact

For challenge questions, post in the Kaggle discussion forum so answers remain visible to every team. Use the team registration form for registration details.