← Back to Design & Engineering
MuM Soft Robotics Visualization
Summer research at TUHH on the MuM soft continuum robot: four IMUs, six tendon servos, micro-ROS firmware on a Teensy 4.1, and a ROS 2 Jazzy digital twin in RViz that tracks the physical arm in real time.
Background
The MuM soft continuum robot at TU Hamburg is a tendon-driven soft arm for manipulation and human-robot interaction research. Continuum structures bend smoothly instead of at discrete joints, which makes them useful in cluttered or human-adjacent settings but harder to model and teleoperate than rigid arms. A practical lab setup needs on-robot sensing and a visualization layer so researchers can see shape and actuator state while the hardware moves.
Over summer 2026 I built an integrated sensing and visualization stack for this platform: firmware that streams IMU and servo data over micro-ROS, plus a ROS 2 Jazzy package that estimates continuum shape and renders a matching digital twin in RViz. The goal was one repo a new lab member could clone, flash, and launch without stitching together separate projects.
Technical architecture
On-robot electronics center on a Teensy 4.1 running integrated PlatformIO firmware that reads four LSM6DSV16X IMUs and drives six DS20-style tendon servos. The tip IMU uses on-chip SFLP fusion; segment orientations feed a constant-curvature shape pipeline on the host. A USB serial micro-ROS agent (Docker, Jazzy) bridges the MCU to the ROS graph.
The continuum_twin ROS 2 package publishes shape topics, runs a twin node for RViz visualization, and includes a tuning GUI for six-motor slider control with a zeroing workflow. Launch files cover visualization, servo-only bring-up, and IMU calibration. Shape estimation combines tip fusion with arc and quaternion math; the URDF and RViz config live in the same package so the twin matches the physical robot frame conventions.
Results and iteration
The side-by-side demo shows the physical arm and RViz model tracking together after a short startup calibration hold. Early work focused on reliable micro-ROS memory limits and agent reconnect behavior; later iterations tightened servo PWM mapping, topic naming, and calibration prompts so shape output stayed stable during tendon pulls. Setup is documented in the repo README and docs guides, including firmware flash, agent startup, and colcon build.
Next steps include fuller four-IMU fusion, load-cell integration, current sensing on servos, and tighter coupling with the lab ros-hub orchestration layer.