Attending the 2nd German Robotics Conference (GRC) 2026 in Cologne

Created on February 25, 2026

2026

I’m pleased to share that our paper “A Novel Powered Jaw Exoskeleton to Treat Temporomandibular Disorders: Design and Control Challenges” has been accepted and is presented at the 2nd German Robotics Conference (GRC 2026) in Cologne!

This work presents our novel hybrid jaw exoskeleton design along with a discussion of the key control challenges that must be overcome on the path toward safe, wearable robot-assisted therapy for temporomandibular disorders (TMDs).

Temporomandibular disorders affect approximately $5$–$12\%$ of the global population, severely impairing essential functions like chewing, speaking, and swallowing. Despite clear clinical need, powered jaw exoskeletons remain largely unexplored. To the best of our knowledge, only five have been reported in the literature, each with significant limitations. Our work aims to close this gap by providing a systematic approach, a concrete design, and a roadmap for the associated control challenges.

The proposed exoskeleton follows a hybrid rigid-soft architecture that balances effective force transmission with user safety:

  • Rigid chin cup for precise force application to the mandible via four tendon-driven actuators
  • Compliant soft facial mask (modeled with a nonlinear hyperelastic material) to distribute contact forces and protect sensitive facial structures
  • High-fidelity MuJoCo simulation integrating a 24-muscle biomechanical jaw model with deformable finite element soft-body dynamics for evaluation of kinematic tracking, contact forces, and mask stress/strain

A significant contribution of this paper is the systematic identification and analysis of the control challenges inherent to such a system:

  • Contact-rich, discontinuous dynamics from cable taut/slack transitions and intermittent face–mask contact
  • Partial observability, as mask deformation and contact state are not directly measurable
  • Unilateral tendon actuation requiring explicit tension allocation with non-negativity constraints
  • Safety-critical hard constraints on tendon tension, jaw torque, and interface pressure
  • Long-horizon error compounding in model-based rollouts

To address these challenges, we outline a planned control pipeline combining a learned, deformation-aware dynamics model with latent states (inferred from IMU, tendon force, pressure, and EMG history via a recurrent encoder) with a constrained, differentiable model predictive control (MPC) scheme. The MPC cost weights and constraint margins are tuned end-to-end via reinforcement learning, using the MPC as a safety filter that guarantees constraint satisfaction while adapting to maximize task performance.

You can find the full paper details on my publications page and the associated simulation framework on the project website.