Contact-Rich and Deformable Foot Modeling for Locomotion Control of the Human Musculoskeletal System

Haixin Gong, Chen Zhang, Yanan Sui

TL;DR

This paper introduces a novel, deformable foot model integrated into a human musculoskeletal simulation. By addressing the limitations of rigid foot models, their approach enables more realistic human locomotion simulation and provides valuable insights for humanoid robot design.

What Problem Does This Paper Address?

Simulating realistic human locomotion is crucial for understanding biomechanics and designing advanced humanoid robots. A key challenge lies in accurately modeling the human foot, which acts as the primary interface with the ground. Existing musculoskeletal models often simplify the foot as a rigid structure with limited contact points. This oversimplification leads to unrealistic force distribution, insufficient shock absorption, and a limited ability to capture the dynamic, deformable nature of the foot. Consequently, these models struggle to accurately replicate the nuanced interactions that enable stable, efficient, and adaptive human gait. This paper aims to bridge this gap by developing a contact-rich and deformable foot model that enhances the fidelity of human musculoskeletal simulations.

Methods

The authors developed a novel approach that integrates a contact-rich and deformable foot model into an existing musculoskeletal system (MS-Human-700).

  • Deformable Foot Model: The foot model is constructed using a mesh-based geometric representation, incorporating material properties that allow for deformation. This deformable mesh is attached to the rigid musculoskeletal model, acting as a child node in the kinematic tree. To manage computational complexity, vertices are restricted to a single radial translational degree of freedom, and flex edges are constrained to preserve their initial lengths. These flex elements are treated as passive DoFs, allowing natural deformation in response to contact dynamics.
  • Two-Stage Locomotion Control: To tackle the challenges of controlling a high-dimensional system with complex contact dynamics, a two-stage policy training strategy is employed.
  • Motion Retargeting: Real-world human walking data is captured and used to generate SMPL parameters. These parameters are then used to derive joint kinematics for the musculoskeletal model, ensuring the retargeted motion is physically plausible.
  • Reinforcement Learning (RL): * Stage 1 (Rigid Model): An initial policy is trained on the simplified, rigid musculoskeletal model. This stage is computationally efficient and provides a foundational understanding of locomotion control.
  • Stage 2 (Interface-Enhanced Model): The policy trained in Stage 1 is adapted and further trained on the interface-enhanced, deformable foot model. This stage leverages the prior knowledge from the rigid model to guide the agent towards beneficial state-space regions, facilitating learning of more complex contact dynamics.
  • Contact Properties: The paper details the mechanical properties of the flex elements used for the foot, including radius, mass, Poisson's ratio, and Young's modulus. They also highlight the importance of carefully designed solver parameters, such as constraint impedance and reference acceleration, for achieving robust contact realism and compliance.

Datasets

  • Human Walking Data: Real-world human walking sequences were recorded using motion capture and insole sensors to measure plantar pressure distribution. This data was used for motion retargeting and for validating the simulation results.
  • Existing Musculoskeletal Models: The MS-Human-700 model served as the base for the musculoskeletal system.

Key Insights and Ideas

  • Deformable Feet Matter for Realism: The study strongly suggests that modeling the foot's deformable nature is not just about anatomical accuracy but is critical for achieving realistic ground contact dynamics, shock absorption, and energy efficiency in locomotion simulations.
  • Two-Stage Training is Key: Directly training an RL agent on a complex, deformable foot model is computationally prohibitive and unstable. The proposed two-stage approach, leveraging a simpler rigid model first, significantly improves learning efficiency and stability.
  • Contact Richness Enhances Stability: The deformable foot model leads to a larger contact area and more distributed contact forces, which in turn improves balance and robustness, providing a larger stability margin compared to rigid models.
  • Bridging Sim-to-Real: The improved fidelity of the deformable foot model offers a more realistic simulation environment, which can directly inform the design of humanoid robots, potentially reducing the sim-to-real gap.

Key Results

  • Improved Kinematic and Kinetic Fidelity: The interface-enhanced model's simulated joint angles and ground reaction forces more closely matched human experimental data compared to the rigid model.
  • Enhanced Contact Characteristics: The deformable foot model exhibited a broader and more evenly distributed ground reaction force pattern, with a higher contact area and more continuous contact transitions.
  • Reduced Force Peaks: The interface-enhanced model showed a lower peak ground reaction force (876 N) compared to the rigid model (1116 N), indicating more distributed and less impulsive force transmission.
  • Increased Stability: The deformable model demonstrated a 34.3% reduction in the coefficient of variation for velocity and a 21.4% reduction for acceleration, indicating more consistent and stable locomotion patterns.
  • Efficient Energy Transfer: The deformable model exhibited higher kinetic energy and smoother potential energy fluctuations, suggesting a more efficient energy storage-return mechanism, analogous to the human foot's arch.
  • Validation with Human Data: Comparisons with real-world human data showed that the interface-enhanced model's knee and hip joint angles aligned better, and its plantar pressure distribution (heel and forefoot forces) more closely matched human measurements, particularly in capturing gradual load transfer.

Limitations and Future Work

  • Computational Cost: While the two-stage training strategy addresses computational challenges, simulating deformable contact remains inherently more intensive than rigid contact.
  • Generalizability: The current framework is validated for normal walking. Future work could extend this to pathological gait patterns and other complex locomotion tasks.
  • Clinical Applications: The lack of comprehensive physiological data limits further validation for clinical applications.
  • Advancements in Physics Engines: The authors acknowledge that advances in physics engines are needed to further enhance the fidelity of deformable contact modeling.
  • Practical Applications: The insights gained from this research can be translated into practical applications for exoskeleton design and humanoid robotics, aiming for more stable, energy-efficient, and human-like locomotion.

Summary updated

Abstract

The human foot serves as the critical interface between the body and environment during locomotion. Existing musculoskeletal models typically oversimplify foot-ground contact mechanics, limiting their ability to accurately simulate human gait dynamics. We developed a novel contact-rich and deformable model of the human foot integrated within a complete musculoskeletal system that captures the complex biomechanical interactions during walking. To overcome the control challenges inherent in modeling multi-point contacts and deformable material, we developed a two-stage policy training strategy to learn natural walking patterns for this interface-enhanced model. Comparative analysis between our approach and conventional rigid musculoskeletal models demonstrated improvements in kinematic, kinetic, and gait stability metrics. Validation against human subject data confirmed that our simulation closely reproduced real-world biomechanical measurements. This work advances contact-rich interface modeling for human musculoskeletal systems and establishes a robust framework that can be extended to humanoid robotics applications requiring precise foot-ground interaction control.
BibTeX
@inproceedings{gong2025contact,
  author = {Haixin Gong and Chen Zhang and Yanan Sui},
  title = {{Contact-Rich and Deformable Foot Modeling for Locomotion Control of the Human Musculoskeletal System}},
  booktitle = {Humanoids},
  year = {2025},
}

Loading related content...

Figure