IROS 2026 Workshop · Pittsburgh, Pennsylvania
Beyond Exteroception Interoceptive Perception for Resilient Robotics
131 teams · 2,800+ submissions in the Learning IMU Odometry ChallengeWorkshop registration and poster submission details are coming soon.
Abstract
Modern robots increasingly rely on external sensors—cameras, LiDARs, and radars—as their primary perceptual modality. Yet biological organisms evolved a fundamentally different strategy: they first understand their own body through vestibular and proprioceptive feedback before interpreting the external world. This workshop explores internal perception, the use of inertial measurement units (IMUs), joint encoders, force/torque sensors, and other body-mounted proprioceptive sensors, as a primary, not auxiliary, source of perceptual intelligence for resilient robotics.
We argue that robust autonomy demands perception systems that are not only world-facing but also self-aware of their motion, dynamics, and physical state. This is not a metaphorical notion but a principled research direction centered on inertial sensing, proprioception, and their integration with external perception. Topics span learning-based inertial odometry, cross-embodiment proprioceptive motion model, adaptive sensor fusion under degradation, and the emerging role of humanoid robots as testbeds for internal-sensing research. The workshop brings together researchers from state estimation, legged locomotion, inertial navigation, and neuroscience-inspired robotics to define the foundations of this underexplored paradigm. Featuring invited talks, a contributed poster session, a panel discussion, and the inaugural Learning IMU Odometry Challenge, this workshop aims to catalyze a community around instinct-like perception for resilient robots.
Important Dates
Training data, baseline code, and evaluation toolkit released.
Final challenge submission and model weights deadline.
Technical report deadline.
Top teams notified and workshop spotlight invitations issued.
Workshop, challenge spotlight talks, and award announcements.
The final submission deadline is listed in UTC. Live competition rules on Kaggle remain the source of truth.
Workshop Scope
Robots need to understand both the world around them and the state of their own bodies. This workshop examines inertial measurement units, joint encoders, force/torque sensing, and other proprioceptive signals as primary sources of perceptual intelligence—not merely auxiliary inputs to vision and LiDAR pipelines.
The program connects state estimation, legged locomotion, inertial navigation, humanoid robotics, and learning-based perception. Invited talks, challenge spotlights, contributed posters, a panel discussion, and open networking will focus on systems that remain reliable when external sensing is degraded or unavailable.
Topics
- Learning-based inertial odometry and navigation
- IMU foundation models and cross-platform generalization
- Proprioceptive state estimation for legged and humanoid robots
- Multi-IMU fusion and spatial-temporal calibration
- Adaptive sensor fusion under environmental degradation
- Online adaptation and self-supervised learning
- Vestibular and proprioceptive inspiration from neuroscience
- Sim-to-real transfer for internal perception
- Robustness benchmarks and evaluation metrics
- Contact-rich and force-aware state estimation
- Differentiable factor graphs and learned optimization
- Integration with visual and geometric foundation models
Who should attend: researchers, students, and practitioners working in state estimation, inertial navigation, robot learning, legged or humanoid robotics, sensor fusion, and resilient autonomy. No specialized workshop prerequisite is required.
Invited Speakers
The current invited lineup spans locomotion, state estimation, learning, and resilient perception. Additional program updates will be posted as they are finalized.
Professor of Robotics and Perception
University of Zurich
Learning Agile Flight from Vision to Commands: From State Estimation to Stateless Navigation
Associate Professor, Naval Architecture and Marine Engineering and Robotics
University of Michigan
Equivariant Proprioceptive Estimation and Learning for Robotics
Institute Associate Professor
NYU Tandon School of Engineering
Egocentric Experience and Memory for Embodied Spatial Intelligence
Incoming Assistant Professor of Robotics and Artificial Intelligence; Member of Technical Staff
ETH Zurich / Amazon FAR
Talk title to be announced
Member of Technical Staff; Incoming Assistant Professor, Computer Science
Amazon FAR / University of Chicago
Talk title to be announced
Postdoctoral Researcher
GRASP Lab, University of Pennsylvania
Estimating Motion from Canonical, Proprioceptive Representations
Program
| Time | Speaker | Topic |
|---|---|---|
| 8:40 - 9:10 AM | Shibo Zhao Carnegie Mellon University |
Opening Address & Challenge Introduction |
| 9:10 - 9:40 AM | Chen Feng NYU Tandon School of Engineering |
Egocentric Experience and Memory for Embodied Spatial Intelligence Embodied agents must learn not only to perceive the world, but also to organize their egocentric experience into persistent spatial knowledge that supports reasoning and action over time. In this talk, I will present our recent work on learning navigation from large-scale visual experience, building and updating spatial memories in changing environments, and using egocentric representations for downstream interaction. Together, these efforts explore how experience and memory can serve as foundations for robust embodied spatial intelligence. |
| 9:40 - 10:10 AM | Carmelo Sferrazza ETH Zurich / Amazon FAR |
Title to be announced |
| 10:10 - 10:40 AM | Maani Ghaffari University of Michigan (remote) |
Equivariant Proprioceptive Estimation and Learning for Robotics |
| 10:40 - 11:10 AM | Davide Scaramuzza University of Zurich |
Learning Agile Flight from Vision to Commands: From State Estimation to Stateless Navigation |
| 11:10 - 11:40 AM | Social Time & Panel Discussion Invited speakers and attendees |
Before a Robot Can Model the World, Must It Model Itself? World models and vision-language-action policies condition on a body state they cannot produce themselves. Panelists discuss whether a robot's self-model, learned from inertial, proprioceptive, and tactile signals, is a prerequisite for modeling the world, or whether it emerges on its own from end-to-end training at scale. View panel slides → |
| 11:40 - 2:00 PM | Lunch Break — Lunch and networking | |
| 2:00 - 2:30 PM | Yuheng Qiu Amazon FAR (Frontier AI & Robotics) |
Title to be announced |
| 2:30 - 3:00 PM | Daniel Gehrig GRASP Lab, University of Pennsylvania |
Estimating Motion from Canonical, Proprioceptive Representations This talk explores how to leverage the spatial and temporal symmetries of motion to derive canonical representations from inertial sensors. These representations are invariant to changes in orientation and motion speed, simplifying the learning of neural displacement priors and improving their generalization. Drawing on EqNIO and Lie Events, I will show how to design equivariant neural networks and event-driven sampling schemes that not only improve the accuracy and robustness of neural inertial odometry but also reduce the data volume of inertial measurements. |
| 3:00 - 3:30 PM | Coffee Break & Challenge Team Presentations — Top challenge teams present posters and demos, alongside contributed posters and networking | |
| 3:30 - 4:00 PM | Haozhi Qi Amazon FAR / University of Chicago |
Title to be announced |
| 4:00 - 4:30 PM | Wenshan Wang Carnegie Mellon University |
Title to be announced |
| 4:30 - 5:00 PM | Panel Discussion Invited speakers |
Explicit or Implicit? The Future of IMU Learning in Robot Perception Should robots model inertial sensing explicitly, through dedicated and interpretable estimation modules, or implicitly, inside end-to-end learned policies? Panelists discuss what each path means for accuracy, generalization, and resilience when exteroceptive sensing degrades or fails. View panel slides → |
All times are Pittsburgh local time (EDT, UTC−4). The schedule may be adjusted as remaining talks and team presentations are confirmed.
