Loading...
Understand the world. Awaken the intelligence.We score unmanned systems missions physically, down to the rules of engagement.
From the air to underwater, we recreate the operating environments of unmanned systems in a virtual space and score mission success, safety, and compliance with the rules of engagement.
One world. Many weapon systems. Autonomous intelligence.
imSim defense is a simulation designed so that unmanned systems in the air, on the ground, at sea, and underwater can operate together and be validated. It reproduces terrain, weather, and sea states through physics computation, and scores mission success, safety, compliance with the rules of engagement, and maritime collision-avoidance rules, recording each verdict together with its evidence. Reinforcement learning environments can be connected to train the AI of unmanned systems, which is then checked again with the same scorer. Integration with existing tactical simulations through standards (HLA, DIS, CIGI) is planned.
Three areas we add on top of existing tactical simulation.
Existing tactical simulations have evolved over many years on abstract physics models. As unmanned systems multiply and manned-unmanned teaming operations become mainstream, the areas that require precise physics, sensors, and AI training together are growing fast. imSim defense fills that space.
Adds precise 3D physics
We simulate environmental conditions such as rough terrain, weather, and sea states together with the movement of unmanned systems through physics computation. This complements areas that need precise physics, such as rough-terrain driving and maritime navigation.
Adds sensor reproduction
Sensors such as cameras, infrared, depth, and sonar generate their output from the same scene. Conditions such as communication loss and satellite navigation jamming are reflected as well.
Adds an AI training environment
By connecting reinforcement learning environments, you can train the operational AI of unmanned systems (some maritime and ground adversary AI available; imitation learning in development). The trained AI is checked again with the same scorer.
Air, ground, sea, and underwater — we cover all four domains.
Terrain, weather, and sea states are simulated along with the unmanned systems in each domain. Cross-domain cooperative missions are represented in the same environment.
These aerial systems carry out reconnaissance, communications relay, and convoy surveillance missions. Mission routes, warning zones, communication conditions, and camera and infrared sensors are simulated together.
These ground systems carry out reconnaissance, transport, and convoy missions over rough terrain. The interaction between terrain and wheels is reproduced through physics computation, and mission events such as entering contaminated or electronic-warfare zones are recorded.
These surface systems carry out maritime navigation missions that reflect sea states. You can build maritime scenarios in which multiple vessels operate together.
These underwater systems carry out detection and approach missions near underwater structures. Sensor output, including sonar detection, is simulated.
Compose · connect · run · score · review
Compose a scenario, connect the AI and run it, then score the results and compile them into an after-action review.
Mission scenarios (terrain, weather, sea state, communication loss, satellite navigation jamming) are composed in a scenario language. Your AI or our own in-house adversary AI is connected and run in physics simulation, and verdicts are given on mission success, safety, and compliance with the rules of engagement. The verdict evidence and event records are compiled into an after-action review report. Integration with real equipment is planned.
5 core capabilities
High-fidelity 3D environment
Rough terrain, weather, and sea states are reproduced through physics computation, and air, ground, sea, and underwater missions run in the same simulation environment.
Sensor reproduction
Camera, infrared, depth, and sonar sensor outputs are generated from the same scene, with communication loss and satellite navigation jamming conditions reflected.
Rules-of-engagement scoring
Engagement of protected targets, firing without designation, and the grounds for firing (whether self-defense holds) are scored as verdict items, and the verdict evidence is recorded event by event. Target classification is defined by the scenario; an unmanned system's target perception is not yet a scored item.
Mission scope
Engagement and firepower, countering explosives, landmines, and sea mines, naval vessel and maritime defense, special ground missions, and surveillance and reconnaissance are tested as scenarios. Scenarios on real Korean terrain and waters are also available.
AI training environment
Reinforcement learning environments are connected to train the AI of unmanned systems (some maritime and ground adversary AI available; imitation learning in development), and the trained AI is checked with the same scorer.
We first test whether the scorer can tell a violation apart.
We ran the four encounter types of the collision regulations (COLREGs) under identical conditions, once avoiding by the rules and once breaking them. Scoring is trustworthy only if the vessel that follows the rules passes and the one that breaks them fails on that rule's item.
8
tests — 4 encounters × comply · violate
24 / 24
3 repeats per test — all with the expected verdict
8 / 8
re-run in a different execution environment — same verdict
Runs 2026-10-10 ~ 11 · PAVS military profile v0.6 · unmanned surface vessel
Head-on
COLREGs Rule 14
Two vessels meeting head-on each alter course to starboard.
Crossing
COLREGs Rules 15 · 16
The vessel with the other on her starboard side keeps out of the way and avoids crossing ahead.
Overtaking
COLREGs Rules 13 · 16
The overtaking vessel keeps out of the way with ample distance.
Stand-on
COLREGs Rule 17
While the give-way vessel acts, the stand-on vessel keeps her course and speed.
The head-on and crossing violation tests fail even though the vessels pass more than 250 m apart. The avoidance went the opposite way to the rule — a violation that distance alone does not reveal.
Figures = recorded tracks of the first of three repeats per test (simulation engine record · 2 s interval · north up). Both figures of an encounter share one scale.
- Self-conducted test — not a third-party assessment.
- Vessel motion follows scenario commands. The test checks whether the scorer separates compliance from violation, not an AI's piloting ability.
- Simulation verdicts — correlation with real-vessel trials is not yet established.
- Verdict thresholds (risk closest-approach distance, safe distance, etc.) are our own settings.
- Test items and limits — PAVS military profile v0.6 →
Built on open formats.
Open formats keep it compatible with other tools. Integration with defense simulation standards is planned. Tool names indicate what we use or integrate with and are not a performance comparison.
Technologies used
Standards · integration
Where it fits
It applies at each stage of weapon-system development, validation, and operation.
Four ways to grow together.
We work with weapon developers, operators, defense SIs, academic labs, and integration programs in various collaboration models.
- Solution demonstration
- PoC collaboration
- Site evaluation
- Integration with existing simulations
- Open-format integration
- Defense standard integration (planned)
- Academic-lab research
- Government R&D consortia
- New-domain expansion
- Integration project collaboration
- SI partnerships
- Global collaboration
Let's build the next tactical simulation together.
For solution demos, PoC collaboration, or integration project inquiries, get in touch — our team will connect with you quickly.