Frankz

Demo

The following comparison demonstrates why frankz exists.

Background

Researchers face barriers when controlling Franka robots:

1. Not everyone knows ROS - steep learning curve for researchers focused on algorithms
2. C++ expertise required - libfranka’s C++ API excludes Python-first roboticists
3. Focus on algorithms, not control - researchers want to prototype ideas, not debug middleware
4. Existing solutions fall short:
   • franky-control: Pure Python, but fake waypoints—just discrete point-to-point motions
   • Ruckig Cloud API: Has Python bindings, but generates jerky motions via cloud roundtrips

I built frankz to solve this: smooth trajectory generation + pure Python API + real-time performance.

Contribution

Core Features:

• Dual-space motion planning (joint-space + Cartesian-space)
• TOTG (Time-Optimal Trajectory Generation) for speed
• Ruckig integration for jerk-limited smoothness
• 1 kHz real-time control loop via C++ backend
• Zero-overhead Python bindings with nanobind

Performance:

• <1ms trajectory computation latency
• 100% real-time constraint satisfaction
• Zero-copy Python-C++ data transfer

Developer Experience:

1
from frankz import Robot, Motion
2

3
robot = Robot("172.16.0.2")
4
motion = Motion()
5
motion.joint_p2p(target=[0, -0.785, 0, -2.356, 0, 1.571, 0.785])
6
robot.execute(motion)

References

Code Inspiration:

1. MoveIt2 trajectory_processing module - TOTG implementation patterns
2. Modern Robotics Chapter 9 - Trajectory generation

Technical Stack:

• C++17 - Real-time motion generation backend
• nanobind - Zero-overhead Python bindings
• libfranka - Franka Control Interface (FCI)
• Ruckig - Online trajectory optimization

Things I Learned

Low-Level Control:

• Real-time constraints in robotics (1 kHz loop frequency is non-negotiable)
• Memory management in Python-C++ bindings (avoiding copies at 1000 Hz)

Trajectory Theory:

• Time parameterization of paths (how TOTG works under velocity/acceleration limits)
• Jerk limiting for mechanical longevity (third-order continuity matters)

Software Engineering:

• Modern Python packaging (scikit-build-core + PyPI workflows)
• GitHub Actions for cross-platform C++ builds
• API design balancing simplicity and power

Community Give Back

Since I referenced MoveIt2’s trajectory processing code, I contributed back:

Merged PR: Fixed trajectory constraint handling edge case in trajectory_processing module

This ensures the open-source cycle continues—take from the community, improve it, give back.

Acknowledgments

Supervisor: Prof. Ziqi Wang, HKUST

Lab Resources: Von Neumann Institute, HKUST