Humanoid robot throwing a spinning kick
  1. FCL
  2. News
  3. How to pilot a humanoid fighting robot

In depth

How to pilot a humanoid fighting robot

Teleoperation, VR headset, motion capture, simulation: how a pilot controls a humanoid fighting robot, and how to become an FCL pilot.

A human in headgear and gloves facing a humanoid robot in a boxing ring

In a humanoid robot fight, the machine is only half of the fighter. The other half stands outside the ring: the pilot. The pilot reads the opponent, picks the distance and decides when to strike. This report explains in practical terms how a humanoid fighting robot is piloted today, with which tools, how pilots train and how to become one.

Piloting a robot means teleoperating it

Piloting a fighting robot means teleoperating it: controlling it remotely, in real time. Information flows both ways. Commands go from the pilot to the robot (move forward, turn, strike, cover up). Images come back from the robot to the pilot, either through a direct view from ringside or through the robot's cameras shown in a headset.

A piloted robot is different from a programmed one. A programmed robot replays a sequence prepared in advance, the same every time. A piloted robot reacts to what happens: if the opponent backs off, the pilot steps in; if they open their guard, the pilot strikes. At the Hangzhou tournament on 25 May 2025, each Unitree G1 was controlled remotely, in real time, by a human operator.[5][8]

What the robot does, what the pilot does

No human can command all of a humanoid's motors fast enough: the work is shared.

The robot takes care of its balance, where its feet land, how it absorbs a hit and how it gets up after a fall. These reflexes are learned beforehand, often in simulation through reinforcement learning, then loaded onto the robot. At the Hangzhou tournament, robots knocked to the floor got back up on their own in about five seconds in most bouts.[5]

The pilot takes care of everything that makes a fight: distance, guard, rhythm, choice of strikes, combinations, feints. The pilot also manages round time and holds the emergency stop.

Three ways to control a humanoid robot

The controller

The pilot steers the footwork with the joysticks and uses buttons to trigger moves learned in advance: jab, hook, kick, high guard. The upside is responsiveness and precision, as in a fighting video game. The limit: the pilot only has the moves in the library, and the art lies in chaining them at the right moment.

The VR headset and hand tracking

The pilot wears a virtual reality headset. Its cameras track the pilot's hands, and the robot's arms copy those movements. Unitree publishes open-source software of this kind, xr_teleoperate: it works with the Apple Vision Pro, PICO 4 Ultra Enterprise and Meta Quest 3 headsets, with hand tracking or the headset's controllers. In its fullest mode, the robot's upper body follows the pilot, while walking is on the left joystick and turning on the right one, with a speed cap. Version 1.6, released on 29 July 2026, added support for the H2 and R1 models.[1]

In the United States, the company REK has its pilots fight with a VR headset and combat controllers; inside the headset they see what the robot sees, as if from a cockpit.[4]

Whole-body motion capture

The pilot wears sensors on the whole body. Their movements are recorded, adapted to the robot's proportions, then carried out by a controller that keeps the machine balanced. In May 2025, Stanford researchers presented such a system, TWIST, trained with reinforcement learning and imitation.[2] Other work published in May 2026 uses an inertial sensor suit to pilot a G1 with no prior learning.[3] The movement feels as natural as it gets, but not every human move is possible for the robot: its proportions, mass and joints differ from ours.

Latency, the pilot's first opponent

There is always a delay between the pilot's move and the robot's: the time to capture the movement, send it, compute it and have the motors execute it. In a fight, that delay matters. A strike thrown too late hits a guard that is already closed; a slow dodge dodges nothing.

REK says it can make up for part of this latency in software, without publishing figures, and its chief technology officer previously worked on VR latency at Oculus.[4] On the pilot's side, the answer is anticipation: like an online gamer who learns to play with slight lag, the pilot learns to throw strikes a beat earlier.

Training in simulation

A humanoid robot is expensive and wears with every fall. Before touching the real machine, much of the work is therefore done in simulation: a digital twin of the robot, subject to the same physics, on which a move can be repeated thousands of times, falls cost nothing and combinations can be tested.

FCL is developing an open-source project of this kind, fcl-t800-sim, published on GitHub under the MIT licence. It brings together a simulation environment for the T800 robot, a library of fighting moves (footwork, guard, strikes, evasions, falling and getting up) and technical notes. The goal: refine these moves in simulation, then transfer them to the real robot, piloted live by a human. The project is in development and open to engineers, robotics students and pilots.[7]

For a pilot, simulation is also a way to learn the machine: its reach, its inertia, how long it takes to get back into guard, the positions where it risks losing balance.

What makes a good pilot

Piloting a fighting robot does not require coding skills. The qualities that matter come from elsewhere:

  • Reading the fight, from combat sports: managing distance, reading a guard, feeling the moment.
  • Reflexes and precision, from gaming: reacting fast, dosing a command, staying clear-headed under pressure.
  • Composure and presence, from performing: owning a room and a camera.
  • Respect for the machine: knowing its limits, not forcing a joint, knowing when to back off.

Piloting is a skill in its own right. In February 2026 in San Francisco, a 13-year-old pilot beat a former professional MMA fighter in a bout between two Unitree G1s piloted in virtual reality.[6]

How to become an FCL pilot

FCL, the French humanoid robot fighting league, is recruiting the pilots of its 2027 season. The process has four stages:

  1. Application through the form on the become a robot pilot page: who you are, your experience, your social channels if you have any.
  2. Selection: a first conversation, then a trial session at the controls of a robot.
  3. Training on the league's robots, supplied and maintained by the league.
  4. Season: undercard bouts, standings, then the main event for the best, with the FCL French champion's title at stake.

You don't need to own a robot or know how to code: programming and maintenance are handled by the league's technical team. The season starts with FCL 01 in Paris in March 2027; the full calendar is on the robot fighting in France page, and how a fight unfolds in our guide to humanoid robot fighting.

Piloting is not only for competition. When a company rents a robot for an event, an FCL operator pilots it live from start to finish: that is how the league's humanoid robot rental works. For the worldwide picture, also read Humanoid robot fighting: where do things stand in 2026? and, on the hardware side, Which humanoid robots can fight?

Frequently asked questions

Do you need to know how to code to pilot a fighting robot?

No. The pilot controls the robot in real time. At FCL, programming and maintenance are handled by the league's technical team.

What equipment does a robot pilot use?

Depending on the format: a controller, a virtual reality headset with hand tracking, or a motion capture suit. In every case, the pilot has an emergency stop.

Can the robot fight without a pilot?

The robot handles its own balance and can get back up, but fighting decisions (distance, strikes, evasions) belong to the pilot. In FCL fights, every robot is piloted live by a human.

How do you become a fighting robot pilot in France?

By applying on the “become a robot pilot” page of fclbots.com. FCL selects its pilots after a conversation and a trial session, then trains them on the league's robots.

Sources

  1. Unitree, official xr_teleoperate repository (GitHub), version 1.6 of 29/07/2026
  2. Ze and co-authors, “TWIST: Teleoperated Whole-Body Imitation System”, arXiv, 05/05/2025
  3. Research paper on teleoperating a G1 with an inertial suit, arXiv, May 2026
  4. eWeek, 15/08/2025
  5. Yicai Global, 26/05/2025
  6. eWeek, 18/02/2026
  7. FCL BOTS, fcl-t800-sim repository (GitHub)
  8. eHangzhou (Hangzhou city), from Global Times, 26/05/2025

Also read: everything about robot fighting in France, the league and its calendar.

Free guide

Download the free guide

Planning an event with a humanoid robot: budget, space, safety, running order and booking checklist. 8 pages, free.

Get the guide