← Paper Digest

SG-ODOM: Event-Driven Proprioceptive State Estimation with Stability-Gated Contact Updates for Legged Robots

Francesco Iotti, Chiyun Noh, Franco Angelini, Manolo Garabini, Ayoung Kim, Maurice Fallon, Frank Dellaert

November 2026

TL;DR SG-ODOM is leg odometry that applies kinematic corrections only after a foot's contact has settled, plus timeout updates that cap IMU-only stretches at about 0.1 s. Its smoother version has the lowest average drift on ANYmal, Spot and a Unitree G1, 0.157 m against 0.208 m for ANYbotics' onboard estimator on seven GrandTour missions.

What they do

Proprioceptive odometry for legged robots fuses IMU propagation with leg kinematics and assumes a foot in contact is stationary. Right after touchdown, though, a foot can still bounce or settle, and repeated corrections within one contact reuse strongly correlated measurements. SG-ODOM changes when kinematic corrections happen: only once a contact has stabilized, and then at a bounded, lower rate. It is tested on Spot, ANYmal D and a Unitree G1 humanoid.

How it works

  • Stability gate. Ground reaction force is estimated from joint-torque residuals. A contact counts only once the vertical force is large enough and steady for a set time, so the delay after impact varies per step.
  • Event-driven updates. A validated touchdown triggers an immediate correction. Otherwise a timeout update fires once IMU-only propagation reaches a maximum interval (0.1 s). Contacts are still classified at full rate.
  • Two back ends. An invariant EKF (SG-ODOM-InEKF) and a fixed-lag smoother (SG-ODOM-FL), both in GTSAM, share one robot-specific front end.

Results

  • Error is lowest for update intervals of 0.05-0.10 s. On ANYmal the smoother’s translational error rises from 0.157 m to 0.418 m at 0.30 s and to 0.406 m at 0.01 s.
  • Stability gating beats a force threshold and, where available, the manufacturer’s contact state for both estimators on all three robots (ANYmal smoother: 0.157 m vs 0.259 m and 0.210 m).
  • ANYmal (seven GrandTour missions): SG-ODOM-FL averages 0.157 m, against 0.208 m for the onboard ANYbotics estimator, and is best on 5 of 7. Spot: 0.185 m vs 0.226 m for Pronto. G1: 0.109 m, about half of Pronto and Unitree’s odometry.

Limitations

The G1 reference comes from LiDAR-inertial odometry rather than absolute ground truth, and the G1 only walked on paved concrete, where gating helped least. The smoother’s IMU bias parameters were chosen empirically per robot, without calibration data. The filter version averages 0.252 m on ANYmal, worse than the ANYbotics estimator.

Videos

Full video (YouTube)

Figures

From Figure 4: The three test platforms: Boston Dynamics Spot (left), ANYbotics ANYmal D (center) and the Unitree G1 ...
From Figure 4: The three test platforms: Boston Dynamics Spot (left), ANYbotics ANYmal D (center) and the Unitree G1 humanoid (right).
From Figure 1: Event-based contact updates for a trotting quadruped. Stable touchdowns trigger immediate updates, lat...
From Figure 1: Event-based contact updates for a trotting quadruped. Stable touchdowns trigger immediate updates, later timeout updates bound IMU-only propagation, and each update resets the interval, shown for limits of 100 ms and 200 ms.
From Figure 5: Contact classification for the left-front foot of ANYmal while trotting: foot height, the ANYbotics co...
From Figure 5: Contact classification for the left-front foot of ANYmal while trotting: foot height, the ANYbotics contact state against the stability-gated one, and the estimated vertical ground reaction force. Highlighted intervals mark transient samples the default classifier accepts but the gate rejects.
From Figure 6: Unitree G1 on the SLOPE sequence, path from above (top) and height over time (bottom). Against a LiDAR...
From Figure 6: Unitree G1 on the SLOPE sequence, path from above (top) and height over time (bottom). Against a LiDAR-inertial reference, Pronto drifts downward while SG-ODOM-FL follows the elevation more closely.
Abstract

Reliable proprioceptive state estimation for legged robots fuses inertial sensing with kinematic sensing from leg encoders. The latter assumes a foot is stationary during the stance phase; however, physical foot touchdown does not necessarily provide a stationary reference for odometry, particularly during impact and load transfer. Repeatedly applying kinematic corrections throughout the same contact phase can introduce redundant or unreliable information. We present SG-ODOM, a proprioceptive odometry framework that uses stability-gated contact selection to schedule kinematic updates. Following physical touchdown, a stability gate accepts the contact for state estimation only after the foot reaches a sufficiently stable support condition. The resulting validated touchdown event triggers an estimator correction, while additional timeout updates bound the maximum duration of uninterrupted IMU propagation. The framework is evaluated on multiple legged platforms, both quadrupedal and humanoid robots, against representative state-of-the-art proprioceptive estimators. Across the evaluated sequences, SG-ODOM demonstrates competitive or superior estimation accuracy, reducing Relative Pose Error (RPE) by up to 25% compared to the baseline estimator on outdoor sequences from the GrandTour dataset. Source code: https://github.com/ori-drs/SG-ODOM. Video: https://youtu.be/-583_RyuhqQ

Summaries are written automatically from each paper's text. Figures and videos come from the paper and its project page; all credit goes to the authors.