SG-ODOM: Event-Driven Proprioceptive State Estimation with Stability-Gated Contact Updates for Legged Robots
Authors: Francesco Iotti, Chiyun Noh, Franco Angelini, Manolo Garabini, Ayoung Kim, Maurice Fallon, Frank Dellaert
Organizations: Centro di Ricerca “Enrico Piaggio” and Dipartimento di Ingegneria dell’Informazione, Università di Pisa, Pisa, Italy. · Department of Mechanical Engineering, Seoul National University, Seoul, Republic of Korea. · Dept. of Engineering Science, University of Oxford, Oxford, UK. · College of Computing, Georgia Inst. of Technology, Atlanta, GA, USA.
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
Figures & tables
Fig. 1 : Illustration of the event-based contact update for a quadruped trot gait. Stable touchdown events initiate immediate contact-triggered updates. Later timeout-triggered updates bound IMU-only propagation (below Δtmax ). Each update resets the dead-reckoning interval, resulting in event-driven rather than fixed-rate updates. This is illustrated here for two different thresholds of 100 ms and 200 ms.
Fig. 2 : Overview of the proposed modular proprioceptive odometry framework. The robot-specific front end converts platform-dependent sensing and kinematics into common interfaces, which are processed by a shared estimation core supporting InEKF and fixed-lag smoothing.
Fig. 3 : The highlighted foot is shown over the course of pre-touchdown swing, impact and transient loading, stable contact, and liftoff. The vertical GRF illustrates how contact activation is delayed until the loading transient has stabilized, rather than occurring at the earliest instant of physical touchdown.
Fig. 4 : The proposed proprioceptive odometry framework is demonstrated across three legged platforms: Boston Dynamics Spot (left), ANYbotics ANYmal D (center), Unitree G1 humanoid (right).
Δtmax
ANYmal
Spot
Unitree G1
(Tgait=0.65s)
(Tgait=0.70s)
(Tgait=0.85s)
0.30s
0.484
0.512
0.131
0.20s
0.265
0.227
0.119
0.10s
0.252
0.210
0.115
0.05s
0.250
0.211
0.114
0.01s
0.255
0.213
0.122
TABLE I : Effect of the maximum interval between kinematic corrections on translational RPE for SG-ODOM-InEKF. RPE t is reported in meters, and Tgait denotes the average gait period for each platform.
Δtmax
ANYmal
Spot
Unitree G1
(Tgait=0.65s)
(Tgait=0.70s)
(Tgait=0.85s)
0.30s
0.418
0.504
0.121
0.20s
0.202
0.215
0.114
0.10s
0.157
0.185
0.109
0.05s
0.157
0.197
0.110
0.01s
0.406
0.398
0.118
TABLE II : Effect of the maximum interval between kinematic corrections on translational RPE for SG-ODOM-FL. RPE t is reported in meters, and Tgait denotes the average gait period for each platform.
Fig. 5 : Contact classification for the left-front foot of ANYmal while trotting. Top to bottom: world-frame foot height, contact classification by ANYbotics (red) classifier and the implemented one γi(t) , and estimated vertical GRF . Highlighted intervals mark samples accepted by the default classifier but rejected by the γi(t) during contact transients.
Contact strategy
ANYmal
Spot
Unitree G1
Manufacturer-provided
0.291
0.224
–
Force threshold
0.307
0.228
0.122
Stability-gated
0.252
0.210
0.115
TABLE III : Effect of the contact strategy on translational RPE for SG-ODOM-InEKF. RPE t is reported in meters. Bold values indicate the best results.
Contact strategy
ANYmal
Spot
Unitree G1
Manufacturer-provided
0.210
0.203
–
Force threshold
0.259
0.190
0.113
Stability-gated
0.157
0.185
0.109
TABLE IV : Effect of the contact strategy on translational RPE for SG-ODOM-FL. RPE t is reported in meters. Bold values indicate the best results.
ALB-3
GRI-1
HEAP-1
LEICA-1
PIL-2
SBB-1
SNOW-3
Average
MUSE [ 25 ]
0.418
0.395
0.175
0.223
0.183
0.160
0.511
0.295
IEKF [ 18 ]
0.398
0.401
0.181
0.194
0.159
0.360
0.512
0.315
IS [ 31 ]
0.397
0.403
0.182
0.193
0.158
0.363
0.515
0.316
ANYbotics Estimator
0.366
0.164
0.144
0.155
0.179
0.293
0.156
0.208
SG-ODOM-InEKF
0.308
0.158
0.239
0.587
0.155
0.149
0.167
0.252
SG-ODOM-FL
0.263
0.169
0.153
0.131
0.127
0.121
0.133
0.157
TABLE V : Translational RPE on ANYmal using missions from the GrandTour dataset [ 14 ] . RPE t is evaluated over 1m path-length intervals and reported in meters. The final column reports the average across the seven missions. Bold and underlined values indicate the best and second-best results, respectively.
Downstair
Overpass
MoCap-E
MoCap-H
Average
RPE t
RPE r
RPE t
RPE r
RPE t
RPE r
RPE t
RPE r
RPE t
RPE r
Pronto [ 5 ]
0.196
0.925
0.180
0.804
0.264
0.652
0.263
0.887
0.226
0.817
MUSE [ 25 ]
0.182
1.095
0.191
0.972
0.252
1.831
0.267
1.733
0.223
1.408
SG-ODOM-InEKF
0.175
0.791
0.125
0.930
0.252
0.648
0.288
0.888
0.210
0.814
SG-ODOM-FL
0.117
0.813
0.118
0.930
0.250
0.674
0.257
0.869
0.185
0.821
TABLE VI : Relative pose errors on Spot using sequences from the GaRLILEO dataset [ 26 ] . RPE t and RPE r are evaluated over 1m path-length intervals and reported in meters and degrees, respectively. The final column reports the average across all sequences. Bold and underlined values indicate the best and second-best results, respectively.
QUAD
SLOPE
Average
RPE t
RPE r
RPE t
RPE r
RPE t
RPE r
Pronto [ 5 ]
0.220
6.032
0.203
5.156
0.212
5.594
Unitree Odometry
0.263
6.107
0.232
5.773
0.248
5.940
SG-ODOM-InEKF
0.116
5.412
0.114
2.386
0.115
3.899
SG-ODOM-FL
0.112
5.391
0.106
2.328
0.109
3.859
TABLE VII : Relative pose errors on Unitree G1 using self-collected sequences. RPE is evaluated over 1m path-length intervals and reported in meters and degrees. The final column reports the average across all sequences. Bold and underlined values indicate the best and second-best results, respectively.
Fig. 6 : Representative trajectory estimates. Top row shows trajectories in horizontal plane and bottom row shows vertical trajectories. Circles denote start points, and crosses denote end points.
One of the significant challenges in legged robotics is achieving accurate odometry using only onboard proprioceptive sensors. In this study, we present a complete leg odometry pipeline based on an Error-State EKF (ESEKF) that relies exclusively on proprioceptive data: a body fixed IMU, joint encoders, and force sensors, where filter's state is corrected by feet determined to be in a stationary stance. The core of our contribution is fused contact detection and an uncertainty quantification module designed to explicitly identify and reject slippage. This module runs two detectors in parallel for each foot, 1) a debounced, force-based Gaussian Mixture Model (GMM) guided Finite State Machine (FSM) to confirm physical contact, and 2) a kinematic-based Generalized Likelihood Ratio Test (GLRT) on the estimated velocity of the foot. The continuous quality scores from both estimators are fused to detect if the foot is both physically loaded and kinematically stationary and served as an uncertainty signal for each contact. To validate our approach, we collected a multi-modal dataset of 29 sequences spanning diverse indoor and outdoor terrains (e.g., concrete, grass, pebble, and rock) total of 2.4 km long. We benchmarked our approach against both proprioceptive and exteroceptive methods. The results demonstrate our method's efficacy in providing accurate odometry estimates, robustly handling slippage-prone environments. We also share our code and real-time ROS2 package as open-source.
Emre Girgin, Cagri Kilic
Department of Aerospace Engineering, Embry-Riddle Aeronautical University, Daytona Beach, FL 32114, USA.
Legged robots carry an IMU, but the inertial solution drifts because consumer-grade IMUs are noisy. However, the feet create intermittent contacts with the environment that can be used to mitigate that drift. This report develops a sequence of increasingly expressive legged robot state estimators that leverage this. In all cases, the floating-base state comprises attitude, position, velocity, and IMU biases. To model foot contacts, we start from the contact-aided invariant EKF of Hartley et al., albeit at a reduced contact update rate. This is then augmented by replacing the measurement update by a small factor graph. Finally, we turn the same factors into a fixed-lag smoother with contact-episode footholds, with and without an evolving IMU bias. To facilitate reproducibility and further research in proprioceptive legged odometry, all four variants are available in GTSAM (Dellaert et. al), and we additionally provide a ROS2-compatible implementation.
State estimation for legged robots remains challenging because legged odometry generally suffers from limited observability and therefore depends critically on measurement constraints to suppress drift. When exteroceptive sensors are unreliable or degraded, such constraints are mainly derived from proprioceptive measurements, particularly contact-related leg kinematics information. However, most existing proprioceptive odometry methods rely on an idealized point-contact assumption, which is often violated during real locomotion. Consequently, the effectiveness of proprioceptive constraints may be significantly reduced, resulting in degraded estimation accuracy. To address these limitations, we propose an interacting multiple model (IMM)-based proprioceptive odometry framework for legged robots. By incorporating multiple contact hypotheses within a unified probabilistic framework, the proposed method enables online mode switching and probabilistic fusion under varying contact conditions. Extensive simulations and real-world experiments demonstrate that the proposed method achieves superior pose estimation accuracy over state-of-the-art methods while maintaining comparable computational efficiency.
Wanlei Li, Zichang Chen, Shilei Li +2
The authors are with the Shenzhen Key Lab for Advanced Motion Control Technology and Modern Automation Equipment and the Guangdong Provincial Key Laboratory of Intelligent Morphing Mechanisms and Adaptive Robots, School of Intelligence Science and Engineering, College of Artificial Intelligence, Harbin Institute of Technology Shenzhen, Shenzhen 518055, China.