A platform taking shape
CAD is complete. Physical development is beginning, and a concept multi-robot visualization is in progress.
AUTONOMOUS SYSTEMS / PLANETARY CONSTRUCTION
One platform.
Many robots.
A shared mission.
Exploring how decentralized robot swarms can build infrastructure together — and adapt when the mission changes.
A MONASH DEEPNEURON RESEARCH PROJECTSCROLL TO EXPLORE ↓
MISSION CONTEXTLunar infrastructure
RESEARCH FOCUSAdaptive role allocation
APPROACHDecentralized coordination
DEVELOPMENT STAGEPrototype
01 / THE PROJECTBUILDING THE FOUNDATIONS
Infrastructure starts
with coordination.
EMERGENCE investigates a decentralized, homogeneous construction swarm for future lunar infrastructure.
One robot architecture. Multiple construction roles. The project brings excavation, transport, deposition and compaction into a shared workflow, with roads, landing pads and berms as target applications.
Our research focus is the integration of learned task allocation with that workflow — and, over time, validation on embodied robots.
A coordinated workflow
Develop and evaluate shared-policy learning for adaptive role allocation across construction tasks.
Resilient infrastructure
Explore multi-robot construction, ISRU integration, maintenance and infrastructure expansion.
02 / SYSTEM ARCHITECTUREFROM DECISIONS TO ACTUATION
Designed as one
connected system.
The platform combines perception, onboard compute and low-level control with a planned learned coordination layer.
PLANNED AUTONOMY
Shared-policy learning
Centralized training · decentralized execution
POLICY↓
01 / SENSE
→Perception
OAK-D Lite
360° 2D LiDAR baseline
02 / DECIDE
→Onboard compute
Jetson Xavier NX
Current compute platform
03 / CONTROL
Low-level control
STM32F446RE
Robot actuation
DOCUMENTED SYSTEM DESIGN · INTEGRATION AND VALIDATION AHEAD
THE ROBOT PLATFORM
Shared hardware.
Flexible roles.
A homogeneous swarm uses the same robot architecture across the construction workflow. The platform is designed around mobility, material handling and compaction.
- Mobility
- 6-wheel rocker-bogie
- Manipulation
- 3-DOF arm
- Tooling
- Scoop / compactor
- Compaction concept
- PACT-inspired dual-eccentric vibratory plate
03 / AUTONOMY & RESEARCHPLANNED CAPABILITY
THE CENTRAL RESEARCH QUESTION
When conditions change,
can the swarm
change with them?
How can robots adapt their roles under failures and changing resource distribution, while continuing a shared construction mission?
Explore the validation approach ↗EMERGENCE / SYSTEM VISUALIZATION16:9
MEDIA PENDING
LEARNING APPROACHShared-policy MARL
PROPOSED ALGORITHM PATHPPO → MAPPO
TRAINING FRAMEWORKCTDECentralized training, decentralized execution
04 / CONSTRUCTION WORKFLOWTARGET TASKS · PLANNED VALIDATION
Four tasks.
One construction cycle.
The intended workflow connects material movement with surface preparation. These are target capabilities, with physical testing still ahead.
Excavation
Collect material using the arm and scoop.
Transport
Move material between collection and build areas.
Deposition
Place material to form the intended structure.
Compaction
Investigate vibratory compaction for surface preparation.
MISSION APPLICATIONSTarget infrastructure
01 Roads02 Landing pads03 Berms
05 / DEVELOPMENT ROADMAPA MULTI-YEAR RESEARCH PLATFORM
Built in stages.
Validated at every step.
The roadmap separates the project’s present maturity from its proposed next steps and longer-term research direction. Stages are not dated commitments.
CURRENTFrom digital design
From digital design
to physical development.
- ✓CAD complete
The robot architecture has reached completed CAD design.
- ↗Physical development beginning
The project is entering hardware development.
- ↗Single-robot RL policy training
Training a reinforcement learning policy to control a single robot.
- ↗Concept visualization in progress
A multi-robot visualization communicates the intended system; it is not autonomous-results evidence.
PLANNEDConnect intelligence
Connect intelligence
with physical systems.
- 01AI + hardware integration
Deploy the trained RL policy on a single robot for physical testing.
- 02Simulation and learning
Develop toward multi-agent environments and shared-policy learning using PettingZoo and MARL.
- 03Fleet development
Produce 2–3 additional robot units for physical multi-robot testing.
LONG-TERM RESEARCHA platform for resilient
A platform for resilient
construction research.
- 01Multi-robot construction and resilience
Explore coordinated workflows under failures and changing resources.
- 02ISRU integration
Investigate the connection to in-situ resource utilization.
- 03Maintenance and expansion
Extend the research direction toward maintaining and expanding infrastructure.
06 / TECHNICAL STACKHARDWARE & DEVELOPMENT PATH
Grounded in
engineering.
The documented hardware baseline and proposed simulation stack support a progression from single-agent development to multi-agent research.
ONBOARD COMPUTEJetson Xavier NXCurrent
LOW-LEVEL CONTROLSTM32F446RECurrent
PERCEPTIONOAK-D Lite / 360° 2D LiDARCurrent
INITIAL LEARNINGGymnasium / Stable-Baselines3Current
MULTI-AGENT PATHPettingZoo / MARLLater stage
SIMULATIONGazebo / MuJoCoProposed stack
VISUALIZATIONUnreal EngineVisualization direction
07 / RESEARCH & TESTINGEVIDENCE BEFORE CLAIMS
A clear line between
intent and evidence.
The project’s contribution is being explored through integration and embodied validation of learned allocation within a construction workflow.
Adaptation under disruption
Study role allocation under robot failures and changing resource distribution. Evaluation methods and measured results will be documented as research progresses.
From simulation to embodiment
Connect learning environments with the physical construction workflow. Concept visualization supports communication; it does not establish autonomous performance.
CAD complete.
Testing ahead.
No measured performance results, deployed swarm capability or lunar qualification are claimed at this stage.
08 / PARTNERS & COLLABORATORSBUILD WITH US
Engineering is a collaborative effort.
EMERGENCE is presented for robotics companies, space-industry partners, sponsors, researchers and engineering recruiters interested in the project’s development.
Potential areas for collaboration include hardware development, autonomy research, construction testing and support for the research platform.
Start with the project brief ↗09 / CONTACT & COLLABORATIONMONASH DEEPNEURON
Help shape the
next stage.
Explore the project, discuss a research connection,
or consider supporting its development.