Haptic Teleoperation of High-dimensional Robotic Systems Using a Feedback MPC Framework

Jin Cheng, Firas Abi-Farraj, Farbod Farshidian, Marco Hutter

Abstract

Teleoperating high-degree-of-freedom (DoF) robots, such as legged manipulators, requires a delicate balance between complex dynamic planning and high-frequency responsiveness. Traditional Model Predictive Control (MPC) frameworks often struggle with the low update rates necessitated by their computational complexity, leading to poor haptic transparency and instability during contact. This paper introduces a feedback MPC framework that augments the system state with operator commands to generate a target-dependent control policy. By evaluating this policy at 400 Hz—significantly faster than the 70 Hz optimization loop—the system achieves superior tracking accuracy and stable haptic feedback even during high-impact interactions.

Introduction

Teleoperation of complex robotic systems, such as a quadruped equipped with a robotic arm, presents a significant control challenge. To manage the robot's intricate dynamics and physical constraints, researchers typically rely on Model Predictive Control (MPC). While MPC is adept at "looking ahead" to optimize trajectories, its high computational cost often limits update rates to 10–30 Hz.

For a human operator using a haptic interface, these slow update rates are insufficient. Rapid movements by the operator can render the robot’s current plan obsolete, resulting in significant lag (low transparency) and dangerous oscillations or instabilities when the robot interacts with the environment. This gap between the slow "strategic" planning of the MPC and the fast "reactive" requirements of haptic feedback makes precise remote manipulation difficult to achieve.

Methodology: Feedback MPC

The proposed framework bridges the gap between optimization and interaction through a Feedback MPC approach. Rather than outputting a fixed trajectory, the MPC provides a local control policy—a set of instructions that dictate how the robot should react if its state or the operator's command changes.

State Augmentation and Target-Dependent Policy

The key technical innovation lies in State Augmentation. The researchers incorporate the operator’s desired position ($x_d$) and velocity ($v_d$) directly into the MPC’s state vector. Because the operator's goal is embedded in the optimization problem, the MPC produces a feedback gain matrix ($K$) that specifically accounts for changes in human input. This results in a "target-dependent" policy that remains valid even as the operator moves the haptic device.

Multi-Rate Control Architecture

To ensure high-fidelity performance, the system utilizes a three-tiered control hierarchy:

  1. MPC Planner (70 Hz): Solves the heavy optimization problem to generate the optimal policy and feedback gains.
  2. Whole-Body Controller (400 Hz): Evaluates the latest policy from the MPC to adjust joint configurations instantly based on the current high-speed input from the operator.
  3. Low-Level Controller (2.5 kHz): Translates these high-frequency requests into actual motor torques.

This architecture allows the robot to respond to human movements at 400 Hz, maintaining high transparency despite the underlying optimization running at a lower frequency.

Experimental Setup

The framework was evaluated using a sophisticated hardware stack:

  • Robot: An ANYmal D quadruped robot equipped with a 6-DoF robotic arm.
  • Interface: A Force Dimension Omega 6 haptic device, providing 3-DoF force feedback to the operator.
  • Environment: Tests included free-motion tracking, high-impact contact with rigid objects (wooden blocks), and tasks requiring strict self-collision avoidance (e.g., navigating near the robot's own LiDAR sensors).

Key Results

The Feedback MPC framework was compared against a standard MPC baseline and a feed-forward approach. The results demonstrated significant improvements across all metrics:

  • Tracking Accuracy: In free-motion tests, the proposed method achieved a median tracking error of 0.067 m, a substantial improvement over the 0.235 m error observed in the baseline.
  • Haptic Stability: During contact experiments with a wooden block, the feed-forward method became unstable, resulting in violent vibrations. In contrast, the proposed feedback method remained stable even during high-impact contacts involving forces up to 70 N.
  • Constraint Satisfaction: The system successfully maintained safety constraints, preventing the robotic arm from colliding with the robot's chassis or sensors, even when the operator's commands directed the arm toward those restricted zones.

Discussion and Insights

The primary insight of this work is that high-frequency feedback is essential for haptic transparency, but it does not require re-solving the entire optimization problem at that frequency. By calculating the sensitivity of the optimal plan to the target (the feedback gain), the robot can update its behavior locally and remain "optimal enough" between MPC cycles.

Furthermore, because the feedback policy is derived directly from the MPC—which is aware of the robot's physical limits—the high-speed reactions remain safety-compliant. This ensures that the robot does not violate joint limits or self-collision boundaries in its attempt to follow rapid human inputs.

Limitations and Future Work

While the current implementation significantly advances teleoperation for high-dimensional systems, several areas for improvement remain:

  • Degrees of Freedom: The current study focused on end-effector position (3-DoF). Future iterations will aim to include orientation and angular velocity for full 6-DoF control.
  • Operator Modeling: The model currently assumes zero acceleration for the operator's target velocity within an MPC cycle. Incorporating operator acceleration could further reduce overshoot during jerky movements.
  • Expanded Feedback: Moving from 3-DoF force feedback to full 6-DoF feedback (including torques) will be necessary for more complex manipulation tasks.

Summary updated

Abstract

Model Predictive Control (MPC) schemes have proven their efficiency in controlling high degree-of-freedom (DoF) complex robotic systems. However, they come at a high computational cost and an update rate of about tens of hertz. This relatively slow update rate hinders the possibility of stable haptic teleoperation of such systems since the slow feedback loops can cause instabilities and loss of transparency to the operator. This work presents a novel framework for transparent teleoperation of MPC-controlled complex robotic systems. In particular, we employ a feedback MPC approach [1] and exploit its structure to account for the operator input at a fast rate which is independent of the update rate of the MPC loop itself. We demonstrate our framework on a mobile manipulator platform and show that it significantly improves haptic teleoperation’s transparency and stability. We also highlight that the proposed feedback structure is constraint satisfactory and does not violate any constraints defined in the optimal control problem. To the best of our knowledge, this work is the first realization of the bilateral teleoperation of a legged manipulator using a whole-body MPC framework.
BibTeX
@inproceedings{cheng2022haptic,
  author = {Jin Cheng and Firas Abi-Farraj and Farbod Farshidian and Marco Hutter},
  title = {{Haptic Teleoperation of High-dimensional Robotic Systems Using a Feedback MPC Framework}},
  booktitle = {IEEE/RSJ International Conference on Intelligent Robots and Systems},
  year = {2022},
  url = {https://openreview.net/forum?id=GTI6ibJK4v},
}
Figure