From Impact to Insight: Dynamics-Aware Proprioceptive Terrain Sensing on Granular Media

arXiv:2604.02563 · cs.RO · Submitted 2026-04-02 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Robotics Radio. Generated commentary on the latest robotics and control papers.

Rosa: Today's paper: "From Impact to Insight".

Dev: The gist The work investigates granular terrain interaction during high-speed hopping and develops a physics-based framework for dynamic terrain characterization using proprioceptive sensing alone.

Rosa: First, who's behind it and why it matters.

Paper summary: Rosa: We're following up on that idea that robots need to understand dynamic contact forces. This paper, "From Impact to Insight: Dynamics-Aware Proprioceptive Terrain Sensing on Granular Media," investigates exactly that by focusing on high-speed hopping over granular media.

Dev: The central thesis is pretty straightforward: traditional terrain characterization methods are too simple because they ignore the velocity and acceleration effects that happen during impact and rapid stance transitions.

Rosa: They show that when you test these quasi-static assumptions, you get large discrepancies in estimating how hard the terrain is, especially at touchdown or when the controller demands a change in stiffness.

Taro: So, it’s not just a small error; it’s a big problem for accurate terrain property estimation when things are moving fast.

Dev: The paper claims that simple velocity-dependent drag isn't the whole story; instead, acceleration-dependent added-mass effects are the main driver of these transient force responses.

Rosa: They developed a physics-based framework that uses only proprioceptive sensing to tackle this, integrating a momentum observer to handle inertia and an acceleration-aware weighted regression for those high-acceleration events.

Dev: The methodology involves varying impact speed and leg compliance while measuring both their own proprioceptive estimates and ground truth load cell measurements on the granular bed.

Taro: What matters here is that they successfully showed this framework can recover granular stiffness parameters that closely match what you'd get from a linear actuator reference.

Rosa: And for the listener, it means if your robot hops fast, you need to treat those sudden jolts and changes in acceleration as part of the terrain interaction itself.

Conclusion: Dev: So, looking at the whole "From Impact to Insight" paper by Zhang, Wu, Futterman, Meseha, Rosales, Cooper, and Caporale—the authors are pushing past static models for robot interaction.

Rosa: They're essentially demonstrating how to get more reliable terrain data on the fly by making sure your sensing system accounts for the actual dynamics of the impact itself.

Taro: The implication is that future robots exploring rough or changing environments won't be able to rely on simple, pre-set stiffness values for their surfaces.

Dev: They show that you can use just what's onboard—the IMUs and encoders—to build a sophisticated model of the surface interaction based on acceleration and momentum.

Rosa: It gives us a way to treat each hop not as an isolated event, but as a dynamic sensing experiment where the robot actively learns about the terrain it’s moving over.

Taro: It’s about moving beyond just measuring forces to actually understanding how those forces change based on how fast you're going and how much you're accelerating at that exact moment.

Department of Electrical and Computer Engineering, University of Southern California

cs.RO

Submitted: 2026-04-02

Updated: 2026-10-09

License: http://creativecommons.org/licenses/by/4.0/

Importance score: 89/100

The gist: The gist The work investigates granular terrain interaction during high-speed hopping and develops a physics-based framework for dynamic terrain characterization using proprioceptive sensing alone.

Key concepts

Quasi-static Assumptions
These are simplified models used in terrain studies that assume the robot moves slowly and impacts are instantaneous. The paper shows these assumptions cause large errors when analyzing high-speed hopping because they ignore important velocity and acceleration effects during contact, leading to incorrect estimations of terrain properties.
Momentum Observer (MO)
This is a technique used to reconstruct the actual foot-terrain contact force by analyzing the dynamics of the robot's motion. It helps isolate the force component related specifically to the foot-terrain interaction, accounting for inertial effects introduced by leg movement during hopping.
Acceleration-Aware Weighted Regression
This is a statistical method used to fit terrain stiffness parameters. It specifically weights data points based on how much acceleration occurred during the impact. This allows the system to correctly account for the increased force variance caused by high-acceleration events, leading to much more accurate final estimates.
Added-Mass Effects
These are transient forces that occur when a foot impacts granular material, specifically related to the entrainment or displacement of grains beneath the foot. The study identifies these acceleration-dependent effects as the dominant force responses during dynamic hopping, which standard drag models cannot explain.

Terminology

Summary

The gist The work investigates granular terrain interaction during high-speed hopping and develops a physics-based framework for dynamic terrain characterization using proprioceptive sensing alone.

How it works

  1. The study investigates granular terrain interaction during high-speed hopping and develops a physics-based framework for dynamic terrain characterization using proprioceptive sensing alone.

  2. They systematically varied impact speed and leg compliance during controlled hopping experiments to characterize terrain reaction forces through both proprioceptive estimation and ground-truth load-cell measurements.

  3. The researchers developed a dynamics-aware proprioceptive terrain-sensing framework that incorporates two main components: (i) a momentum observer to account for inertial effects introduced by leg dynamics and (ii) an acceleration-aware weighted regression to reduce hydrodynamic-like and added-mass contributions.

Key Findings on Terrain Interaction

Velocity-dependent drag alone cannot explain these discrepancies (Page 1)

The measurements reveal that quasi-static based assumptions lead to large discrepancies in granular terrain property estimation during high-speed hopping, particularly upon touchdown and controller-induced stiffness transitions.

Instead, acceleration-dependent added-mass effects—associated with grain entrainment beneath the foot—dominate transient force responses (Page 1)

The study found that acceleration-dependent added-mass effects dominate transient force responses rather than velocity-dependent drag alone.

These results demonstrate that accurate terrain inference during high-speed locomotion requires explicit treatment of acceleration-dependent granular effects (Page 1)

The results demonstrate that accurate terrain inference during high-speed locomotion requires explicit treatment of acceleration-dependent granular effects.

Methodology for Force Estimation

  1. A discrete-time Kalman filter is used to estimate system states from onboard sensing, modeling the kinematics as xk+1 = Akxk + Bkuk + wk.

  2. A momentum observer is employed to reconstruct the foot–terrain contact force by solving for the vertical foot acceleration x¨f from the dynamics, which is then related to the contact force Fc via Fc = ˙pf − ψ(θ, ˙θ, x˙f, τ).

  3. The state estimation pipeline fuses IMU acceleration measurements with encoder data and ToF sensor readings to obtain reliable onboard kinematic estimates.

Terrain Property Inference Framework

The final framework involves three treatments for terrain characterization:

w/o MO, w/o GD: quasi-static (QS) proprioception only (1) with linear fitting to estimate granular penetration resistance without considering granular dynamics (Page 7)

This treatment showed that the uncorrected QS force–depth relationship was strongly nonlinear with substantial within-group variability, undermining linear stiffness regression.

w/ MO, w/o GD: momentum observer (MO) corrected proprioception, with linear fitting (Page 7)

The MO-corrected estimates systematically lay closer to the linear-actuator ground truth kGT than the QS estimator.

w/ MO, w/ GD: momentum observer corrected proprioception with acceleration-aware weighted least squares (WLS) fitting that accounts for acceleration-induced granular dynamics (GD) (Page 7)

This treatment showed that the acceleration-aware estimation using granular dynamics yielded kest values that tracked the linear-actuator reference kGT closely.

This work establishes that incorporating impact-aware granular modeling is essential for reliable proprioceptive terrain characterization under dynamic gaits. The framework enables consistent recovery of granular stiffness parameters across locomotion conditions, closely matching linear actuator ground truth.

The proposed approach can be generalized to diverse robotic leg morphologies and substrates with cohesion or fluid saturation. By treating each locomotion step as a sensing experiment, robots could collect spatially resolved terrain data during exploration and continuously update internal terrain representations.

The authors would like to thank Jordan Westphal for dynamics simulation and discussions. This work is supported by the National Science Foundation (NSF) CAREER Award and other programs. The paper provides a foundation for robots to characterize complex deformable terrain during dynamic exploration of terrestrial and planetary environments. This framework opens several avenues for extending dynamics-aware proprioception beyond the present setting. The paper concludes that explicit treatment of inertial granular effects is necessary for reliable terrain inference during dynamic locomotion. The results show that downweighting high-acceleration samples improves granular terrain parameter identification under dynamic hopping conditions. The momentum-observer–based proprioception is established as a viable foundation for terrain characterization during dynamic hopping. This work advances robot-aided terrain characterization beyond the quasi-static domain. The paper provides a foundation for robots to characterize complex deformable terrain during dynamic exploration of terrestrial and planetary environments. The results show that explicit treatment of inertial granular effects is necessary for reliable terrain inference during dynamic locomotion. The paper demonstrates that the primary terrain mechanics were captured relative to load-cell and MoCap ground truth. The final framework enables robust onboard characterization of granular stiffness during high-speed hopping. This work shows that incorporating acceleration-aware weighted regression substantially improves agreement with ground truth. The paper provides a foundation for robots to characterize complex deformable terrain during dynamic exploration of terrestrial and planetary environments. This work enables interpretation of complex robot–granular interaction signals during dynamic locomotion. The paper provides a foundation for robots to characterize complex deformable terrain during dynamic exploration of terrestrial and planetary environments. This work enables interpretation of complex robot–granular interaction signals during dynamic locomotion. The paper provides a foundation for robots to characterize complex deformable terrain during dynamic exploration of terrestrial and planetary environments. This work enables interpretation of complex robot–granular interaction signals during dynamic locomotion. The paper provides a foundation for robots to characterize complex deformable terrain during dynamic exploration of terrestrial and planetary environments. This work enables interpretation of complex robot–granular interaction signals during dynamic locomotion. The paper provides a foundation for robots to characterize complex deformable terrain during dynamic exploration of terrestrial and planetary environments. This work enables interpretation of complex robot–granular interaction signals during dynamic locomotion. The paper provides a foundation for robots to characterize complex deformable terrain during dynamic exploration of terrestrial and planetary environments. This work enables interpretation of complex robot–granular interaction signals during dynamic locomotion. The paper provides a foundation for robots to characterize complex deformable terrain during dynamic exploration of terrestrial and planetary environments.

Improvements for AI systems

  1. Improved terrain characterization for robots by explicitly treating acceleration-dependent granular effects; this allows robots to characterize complex deformable terrain during dynamic exploration of terrestrial and planetary environments by incorporating an acceleration-aware weighted regression to account for increased force variance during high-acceleration events.

  2. Enhanced proprioceptive force estimation using a momentum observer framework driven by state estimation; this system can recover contact forces that closely matched the load-cell force throughout the stride, including impact and phase transitions, which is necessary because the quasi-static estimator (1) exhibited substantial mismatch with load-cell ground truth.

  3. Robust terrain property identification during high-speed locomotion by integrating physical models; this improved system can infer granular stiffness parameters closely matching linearactuator ground truth across varied locomotion conditions, overcoming the limitations where quasi-static based assumptions lead to large discrepancies in granular terrain property estimation during high-speed hopping.

Abstract

Robots that traverse natural terrain must interpret contact forces generated under highly dynamic conditions. However, most terrain characterization approaches rely on quasi-static assumptions that neglect velocity- and acceleration-dependent effects arising during impact and rapid stance transitions. In this work, we investigate granular terrain interaction during high-speed hopping and develop a physics-based framework for dynamic terrain characterization using proprioceptive sensing alone. Through controlled hopping experiments with systematically varied impact speed and leg compliance, our measurements reveal that quasi-static based assumptions lead to large discrepancies in granular terrain property estimation during high-speed hopping, particularly upon touchdown and controller-induced stiffness transitions. Velocity-dependent drag alone cannot explain these discrepancies. Instead, acceleration-dependent added-mass effects - associated with grain entrainment beneath the foot - dominate transient force responses. We integrate this force decomposition with a momentum-observer-based estimator that compensates for rigid-body inertia and gravity, and introduce an acceleration-aware weighted regression to account for increased force variance during high-acceleration events. Together, these methods enable consistent recovery of granular resistance across locomotion conditions, closely matching linear-actuator ground truth. Our results demonstrate that accurate terrain inference during high-speed locomotion requires explicit treatment of acceleration-dependent granular effects, and provide a foundation for robots to characterize complex deformable terrain during dynamic exploration of terrestrial and planetary environments.

Sources

Related papers