Skip to main content

Research · three areas

Research questions, stated before results.

Moonpie Robotics is an emerging research initiative. Each area below lists the questions we intend to study and the tools we are evaluating. None of it is a result yet.

Status key

  • Status: Concept
  • Status: Exploratory
  • Status: Proposed
  • Status: In use

01Research area

Robotics and autonomous systems

How a machine perceives a space, decides where it can safely go, and moves through it — developed and tested in software long before it touches hardware.

Status

Status: Concept

Questions we intend to study

  • Robot perception from cameras and depth sensors
  • Motion planning and collision-free navigation
  • Manipulation: reaching, grasping and placing
  • Simulation-first development and evaluation
  • Robot operating systems and hardware integration
  • Human-supervised autonomy, with a person able to stop or override

Candidate technologies

  • ROS 2
  • Python
  • C++
  • Gazebo
  • OpenCV

Candidate technologies under evaluation — not deployed infrastructure.

The six-axis Manipulator concept moving each joint in turn, from base rotation to flange roll
Fig. 2 · Mechanism study · real-time 3DJoint frames in practice: the Manipulator concept exercises its six axes one at a time, base to flange, each to its limits. The arm is posed by the same inverse kinematics that runs its pick-and-place. A concept model; not a built system.
Fig. 3 — pick and place with a two-finger gripper (illustrative)A side-view line drawing of a two-finger gripper closed on a block, with its two contact points marked, an approach direction, and a curved transfer path to a dashed place position on the same table. Illustrative; not a built system.
Fig. 3 · Illustrative figureManipulation as a plan: approach, a two-point grasp, and a transfer path to the place pose. Illustrative; not a built system.

02Research area

Embodied AI

Connecting perception, reasoning and physical action: AI that is judged not by what it says but by what it reliably and safely does in a space.

Status

Status: Concept

Questions we intend to study

  • Linking perception, reasoning and action in one loop
  • Vision-language-action models as a research direction
  • Task planning: turning an instruction into ordered steps
  • Robot learning within simulated environments
  • Measuring reliability and failure modes, not only success
  • Human oversight and controlled, interruptible execution

Candidate technologies

  • Simulation environments
  • Open research models
  • Evaluation harnesses

Research directions, not deployed systems.

Perception–action loop with human approvalThree boxes in a row — perceive, reason, act — joined by arrows. Beneath them, a line joins perceive and act through a highlighted box labelled human approval: a person approves the plan before it runs. A concept drawing.PERCEIVEREASONACTHUMAN APPROVAL
Fig. 4 · Concept drawingPerceive → reason → act, with a person approving the plan before it runs.
The Strider quadruped concept walking forward, stopping, turning in place and side-stepping
Mechanism study · real-time 3DLocomotion on planted feet: the Strider concept walks, stops, turns in place and side-steps. Each foot is placed by a gait planner and held still while it bears weight; the body moves over it. A concept model; not a built system.

How work would progress

  1. 1Software experimentsPlanners and models evaluated against recorded and synthetic data.
  2. 2Simulated environmentsThe same logic run in physics simulation, with failures logged and studied.
  3. 3Supervised hardware prototypesOnly after simulation results justify it, and only under direct human supervision.

03Research area

Neurotechnology and human–machine interaction

Early, non-invasive research into how people and machines can communicate — starting with signal visualisation and accessibility-oriented interfaces, not medical devices.

Status

Status: Exploratory

Questions we intend to study

  • Non-invasive EEG research using public datasets first
  • Brain-signal visualisation and exploration tools
  • Brain–computer interface concepts and their limits
  • Human–machine interaction design
  • Signal processing and machine-learning experimentation
  • Accessibility-oriented interfaces

Candidate technologies

  • Public EEG datasets
  • Signal-processing libraries
  • Python

Non-invasive research only. Nothing here is a medical device, a diagnostic tool or an implantable system.

Read the full roadmap and safeguards

Stages, gates, consent and regulation are on /robotics/neurotechnology.

Stage 01
Contour00:00
Stage 02
Scan00:01.2
Stage 03
Read00:05.2
Source
1280 × 72030fps · 0.35MB

Still frame

Concept visualizationSignal waveforms around a contour head, animated. Not real brain data, not a device, not a photograph.

04Technology under evaluation

What we use, and what we are only considering.

Frontend and visualisationStatus: In use

What these pages are built with, including the sandbox renderer.

Next.js · React · TypeScript · Canvas 2D

Simulation and robotics middlewareStatus: Proposed

Candidates for moving beyond the in-browser sandbox.

ROS 2 · Gazebo · Python · C++

Perception and computer visionStatus: Proposed

Candidates for camera and depth processing experiments.

OpenCV · Depth sensing · Recorded datasets

AI models and task planningStatus: Proposed

Open models evaluated in simulation, with human plan approval.

Task decomposition · VLA research · Evaluation harnesses

Non-invasive biosignal experimentationStatus: Concept

Public datasets first. No live capture, no personal data.

Public EEG datasets · Signal processing

Safety, evaluation and observabilityStatus: Proposed

Logging, repeatable test scenarios and stop controls, defined before any hardware.

Deterministic runs · Failure logging · Human override