MALONI Humanoid: Czech conversation and robot vision in MuJoCo

What is MALONI Humanoid and how does it build on Unitree G1?

The previous Unitree G1 case study introduced the virtual laboratory, walking and assisted object transfer. According to the author, the new stage with an Aimoga Mornine model connects manipulation with conversation and perception: the humanoid completes a task, prepares for conversation and, on request, obtains fresh visual context or performs a specific skill. The purpose is to explore the continuity of movement, perception and human interaction in one environment.

Video from the new development stage

Object transfer and faster task completion

The starting scenario is moving a Buddha figurine onto a table. The humanoid approaches the object, grasps it with assistance, carries it and places it down. After checking stable placement, it steps back, completes a six-second dance and prepares for conversation. Manipulation and stabilisation use simulation assistance mechanisms.

According to internal measurements, the fast profile reduced mean total task time from 97.90 to 79.42 seconds of simulation time. The saving is 18.48 seconds, or 18.88%. The reaching and grasping phase was reduced by 22.09%. This compares profiles of the given simulation task; it is not a measure of physical robot performance or cloud response time.

How does Czech conversation with the humanoid work?

Once the task is complete, the user can start a Czech voice conversation. The panel provides live transcription, typed messages, microphone muting and disconnection. The humanoid accompanies its replies with gentle arm and torso movements.

Conversation uses data directly from the simulation state: the positions of the Buddha and table, whether the object is attached, its stable placement and task duration. These data allow it to describe the actual sequence. They are distinct from visual recognition through a camera: knowing simulation coordinates does not itself prove that the robot recognised the object in an image.

How does the humanoid use its camera and look around?

The camera is mounted on the robot in the MuJoCo simulation. Questions such as “Co vidíš?” (“What do you see?”) or “Kde se nacházíš?” (“Where are you?”) request a new frame and update the visual context before the reply. The response can therefore use a fresh view of the scene.

How are voice commands separated from motion control?

The language model does not set joint positions. Supported Czech phrases select a specific skill, which a local controller executes with limits on speed, joint ranges and obstacle clearance. Network communication and image processing run outside the physics loop, so the simulation continues even when a cloud response is slower.

This separation allows conversation, perception and skill execution to evolve independently. The described limits and interruption with “Stop” are properties of the internal simulation solution; they do not constitute a verified safety system or an emergency stop for a physical robot.

Internal test results in MuJoCo simulation

The following values are taken from Rostislav Sikora’s statement and describe internal testing in an assisted simulation. Times are simulation times; original test reports and a new code revision are not among this study’s public supporting materials.

Developing a platform for Czechia, Slovakia and Poland

This vision points towards MALONI’s own humanoid development platform focused on Czechia, Slovakia and Poland, with regard to applicable EU requirements. Czech conversation is a concrete current part of that direction; it does not substantiate additional languages or operational readiness across the region. The statement expresses a development goal, not certification or demonstrated regulatory compliance.

Here, our own platform means developing scenes, application logic and connections between capabilities. The previous stage used a Unitree G1 model and an existing motion policy; according to the author, the new demonstration uses an Aimoga Mornine model. Public revision da5f9b1 still relates to the previous stage. This does not mean proprietary humanoid hardware or independence from all external components.

Common questions about MALONI Humanoid

Is MALONI Humanoid a physical robot?

This case study describes an internal software project in assisted MuJoCo simulation. According to the author, the new footage uses an Aimoga Mornine model. Simulation results alone do not substantiate a real robot’s capabilities or a customer deployment.

What does the 18.88% reduction in task time mean?

According to internal measurements reported by Rostislav Sikora, the mean duration of the same simulated task fell from 97.90 to 79.42 seconds. Each profile completed ten out of ten validation runs. The figure expresses a relative saving in simulation time, not a demonstrated productivity improvement for a physical robot.

Does the study demonstrate certification or EU compliance?

No. The study documents the direction of internal development as reported by the author, not certification or demonstrated regulatory compliance. Physical robot safety and the requirements for a specific future application will need separate assessment and verification.

What needs to be verified next

The current result provides a practical foundation for further development: completing a manipulation task, moving into conversation, obtaining a fresh view of the surroundings and performing a supported skill. Transfer to a real robot will require separate development, assessment of the specific application and physical verification. This study does not describe a customer installation.

The supplied video and images show the environment of the new development stage. Numerical results and features not directly demonstrated in the footage remain attributed to internal testing. Have a specific development project in mind? Describe the task, environment and expected outcome using the contact link below.