Dr.Anmar Assets

Simulation-ready surgical robotics assets and contact-driven patient effects for NVIDIA Isaac Sim and Isaac Lab.

Autonomous Rescue OR — multi-arm intervention, tool exchange, patient state, resuscitation, and monitoring in one composable research scene.

Dr.Anmar Assets is the standalone simulation substrate behind Dr.Anmar's surgical autonomy research. It combines OpenUSD instruments, robots, patients, operating-room scenes, interaction frames, and Isaac Lab adapters with patient-effect models whose state can only advance from measured post-physics evidence.

The central design rule is simple:

A policy may request an intervention. It may not write the result.

Bleeding control, perfusion, repair integrity, fluid delivery, ventilation, damage, and rescue reward are derived from contact forces, geometry, attachments, flow, pressure, inventory, dwell, and patient state. This places the learning signal downstream of the physical interaction instead of behind a success flag an agent can exploit.

Research review path: Researcher brief → visual showcase → contact-driven effects → machine-readable rescue contract

Patient effects are scene transitions—not reward labels

The shipped inspection states make the physical outcome boundary visible. At runtime, the environment may reach these states only through current contact, geometry, attachment, flow, pressure, dwell, and physiology evidence.

Uncontrolled vessel	Temporary compression	Retained repair
Flow and blood loss remain active.	Benefit expires when bilateral contact evidence disappears.	Benefit persists only while attachment and integrity remain valid.

Leaking bowel ends	Repaired anastomosis
Alignment alone does not establish integrity.	Repair state is coupled to closure geometry, retention, pressure hold, and leak evidence.

Built as a complete surgical research surface

Procedure-scale systems	Mechanism and patient detail
Oncologic resection cell — three coordinated arms, multimodal sensing, margin-aware tissue state, and specimen handling.	SafePlane dissection — an exploded view of interchangeable traction, hydro, blunt, scissors, sensing, and protected-structure components.
Dynamic abdominal patient — layered wall mechanics, open-abdomen access, organs, vessels, pathology, respiration, and physiology.	Articulated skin stapler — trigger, pusher, magazine state, placement frames, and deployable staple representation.
Adaptive hemostasis — compression, clip, patch, suction, irrigation, and verification modes under one contact-owned effect model.	Perfusion viability — RGB, ICG/NIR, speckle, thermal, oxygenation, ultrasound, and Doppler sensing surfaces.

See the visual showcase for full-resolution views and physical-state inspection assets.

Research highlights

• Autonomous Rescue OR — a multi-station rescue environment with complication detection, resource accounting, tool changes, resuscitation, ventilation, and transition-based reward.
• Dynamic abdominal patient — layered OpenUSD anatomy, scenario variants, deformable wound margins, patient physiology, and contact-coupled tissue response.
• Procedure-specific instruments — exposure, wound preparation, hemostasis, dissection, seal-and-divide, anastomosis, perfusion assessment, oncologic resection, closure, and surgical count assets.
• Dependency-complete assets — primary layers ship with their local geometry, materials, textures, interaction frames, physics profiles, and manifests.
• Isaac Lab adapters — configuration and runtime helpers for Franka, dVRK PSM/ECM, STAR, rigid, articulated, particle, and deformable scenes.
• Imitation-learning contract — transition-aligned robot, contact, vessel, vital-sign, and fluid-balance observations with patient effects excluded from the action space and dataset splits kept at complete-episode boundaries.

Contact-driven effect architecture

flowchart LR
    A["RL policy<br/>motion + intervention intent"] --> B["Isaac Sim / PhysX<br/>articulation + contacts"]
    B --> C["Scene evidence<br/>force, gap, speed, pressure,<br/>flow, attachment, dwell"]
    C --> D["Dr.Anmar effect integrators<br/>conservation + damage + repair"]
    D --> E["Patient state<br/>bleeding, MAP, perfusion,<br/>oxygenation, integrity"]
    E --> F["Transition reward<br/>improvement minus harm"]
    F --> A
    A -. "no outcome write path" .-> C

Loading

See Contact-driven effects for the authority boundary and the patient coupling implemented today.

Evidence ownership at a glance

Intervention	Environment-owned evidence	Patient effect	Failure remains observable
Compress or clip	Bilateral force, symmetry, separation, placement, speed, retained attachment	Transient or retained flow reduction	Release, migration, overload damage, distal perfusion loss
Patch or anastomose	Distributed contact, closure gap, attachment integrity, pressure-hold dwell, leaked particles	Seal or repair integrity	Delamination, residual leak, stenosis, rupture
Infuse	Plunger travel, outlet flow, reservoir mass loss, access attachment, line pressure	Intravascular or extravasated volume	Disconnection, occlusion, overpressure, extravasation
Ventilate	Airway attachment, valve travel, delivered and leaked flow, pressure, oxygen fraction, chest excursion	Ventilation and oxygenation support	Leak, unsafe pressure, inadequate delivery

Catalog

Family	Representative assets	Catalog path
Rescue environment	Autonomous Rescue OR, deformable rescue vessel, resuscitation module	data/Environments/SurgicalAutonomy/AutonomousRescueOR
Patient	Dynamic abdominal patient, layered laparotomy anatomy	data/Props/Patients/DynamicAbdominalPatient
Exposure and preparation	Atraumatic exposure, wound preparation	data/Props/SurgicalExposure, data/Props/SurgicalPreparation
Hemostasis and dissection	Adaptive hemostasis, SafePlane dissection	data/Props/SurgicalHemostasis, data/Props/SurgicalDissection
Division and reconstruction	Adaptive seal/divide, adaptive anastomosis	data/Props/SurgicalDivision, data/Props/SurgicalReconstruction
Assessment and oncology	Perfusion viability, OncoSurgery cell	data/Props/SurgicalAssessment, data/Props/SurgicalOncology
Closure and count	Skin stapler, adhesive, needle/thread, closure robot, sponge	data/Props/SurgicalClosure, data/Props/SurgicalCount
Robot foundations	dVRK PSM/ECM and STAR	data/Robots

The detailed inventory is in Asset catalog.

Quick start

Clone the complete repository so every USD layer keeps its relative dependencies:

git clone https://github.com/Numi2/dr-assets.git
cd dr-assets

Use it as an editable package from an Isaac Lab Python environment:

python -m pip install -e .

Or add the checkout as a Kit extension folder. The repository root is the extension root and contains config/extension.toml.

./isaaclab.sh -p your_script.py \
  --ext-folder /absolute/path/to/dr-assets

Resolve a catalog asset in Python:

from orbit.surgical.assets import asset_path

rescue_or = asset_path("autonomous_rescue_or")
hemostasis_tool = asset_path("adaptive_hemostasis")

The orbit.surgical.assets import namespace is retained for compatibility with Dr.Anmar and ORBIT-Surgical-derived tasks. The distributable project name is dranmar-assets.

For submodule integration, environment overrides, and Isaac for Healthcare catalog interoperation, see Integration.

For a compact map of the research claims, implemented authority boundaries, machine-readable interfaces, and highest-value calibration work, read the researcher brief.

What is calibrated—and what is not

The mechanics and patient-effect pathways are executable research code. Parameter sets currently marked provisional_engineering_seeds are not clinical calibration. Asset presence, a successful simulation run, or an RL score does not establish biological fidelity.

The intended calibration path is measured contact/flow/pressure data → identified parameter set → held-out physical validation. Until that work is complete, this repository must not be described as patient-specific, clinically validated, a medical device, or suitable for patient care.

Read the exact research boundaries.

Project ownership and attribution

Dr.Anmar owns the patient-effect architecture, procedure assets, scene composition, interaction contracts, and research integration in this repository. The compatibility namespace and base dVRK/ECM/STAR configuration are derived from ORBIT-Surgical and retain their BSD-3-Clause notices. Dr.Anmar-authored modules and asset families carry Apache-2.0 notices.

NVIDIA Isaac Sim, Isaac Lab, PhysX, and Isaac for Healthcare are technical foundations or interoperability targets; they are not bundled here, and this repository does not imply NVIDIA endorsement. See Provenance and Notices.

Contributing and citing

Research contributions should improve physical observability, calibration, patient coupling, or asset fidelity—not add policy-writable shortcuts. See CONTRIBUTING.md and CITATION.cff.