General, fast, bounded, and robust robotic motion control.
Thracian builds ARC: the same hardware controller for any robot. It plans motion, stays within limits by construction, and closes the loop in real time. Tested on real hardware, with formal verification coming soon to guarantee safety and completeness.
-
IGeneral
Any robot, without the need for per-limb configuration. Give it a URDF and our ARC controller will derive the rest. This makes it so that swapping out robots is a simple switch rather than days of fixes.
-
IIBounded
By construction, our solver will not return a command that breaks any defined limit. Our math simply prevents it, which leads to fewer incidents.
-
IIIRobust
Our controller finds all valid solutions, and if none exist, it states it plainly. Our controller won't suddenly fail on you near a singularity, and nobody has to figure out why the arm suddenly stopped.
-
IVFast
Fast where it's actually difficult to be fast. IK, the dynamics, and the motor commands that follow all run inside a bound you know before you move, not just the last step in the chain.
What we build
Timing lives in hardware
Logic written straight into a real-time parallel processor drives the motor bus where nothing schedules it and nothing preempts it. A hardware pause line stops the arm whether or not the processor is still working.
The motion system
The motion system hands finished motion down to the fabric and steps back. Solving a pose and driving it, the whole chain from IK down, runs inside a bound, every time. Choosing the best path across a full multi-waypoint trajectory is a different problem: that happens once, ahead of time, and it can take as long as it needs.
Simple Python API
Commands are short and explicit, with reporting that has its own channel. An unreachable pose just comes back as an IK failure, and doesn't execute further.
On the record
- aModeling automatic from a URDF, no per-robot build step
- bSupported limbs 3–7 joint arms
- cInverse kinematics complete: every valid branch found and ranked, so no one has to guess which solution the arm "should" pick
- dTrajectory planning multi-waypoint, optimized jointly in task space
- eTiming time-optimal path parameterization, or an exact duration, so a shift gets planned around a real number
- fDynamics full rigid-body inverse dynamics, continuous feedforward
- gGain scheduling continuous, tracking effective inertia through the motion
- hSelf-collision checked while it plans, so the crash never happens in the first place
- iConstraint enforcement by construction; formal verification in progress. One less category of failure your safety review has to take on faith
- jFault handling bad signals not propagated, so one flaky sensor doesn't take the whole line down with it
- kSafety rating none claimed. Not yet certified as a safety controller
In the field
A four-actuator harvesting arm
Strawberry harvesting is a tough first customer that we're aiming for. The fruit is delicate, the harvesting window is short, and the labour to pick it is disappearing. We're building a demo that runs the controller on a printed arm built from parts anyone can order.
Our demonstration aims to prove one claim: the loop closes on time, every time, reliably on hardware you can hold.
Units on loan to robotics labs
Thracian is opening ARC to university robotics laboratories for short-term evaluation on real hardware.
If you run a lab and have an arm that needs a controller, write to us.