[OVPHYSX] IMU sensor

[AntoineRichard]

Summary

Implement `Imu` and `ImuData` for the OVPhysX backend, satisfying the `BaseImu` and `BaseImuData` contracts. Reports angular velocity and linear acceleration in the sensor's body frame using numerical differentiation of velocities.

Validation: Standalone script — attach IMU to falling/rotating rigid body, verify gravity acceleration and angular velocity.

Scope

• `imu.py` (~300 lines), `imu_data.py` (~80 lines), `kernels.py` (~100 lines)
• Copy PhysX tests with config swap

No OVPhysX API Blockers

All required data (body transforms, velocities, COM) is already available through existing Articulation TensorBindings. No new ovphysx APIs needed.

Dependencies

• Requires PR ovphysx backend integration #4852 merged
• Independent of RigidObject specs

Spec

Full design spec: `docs/superpowers/specs/2026-04-20-ovphysx-imu-sensor-design.md`

Parent issue: #5315

---

📋 Full Design Spec (click to expand)

OVPhysX IMU Sensor — Design Spec

Issue: #5318 — [OVPHYSX] IMU sensor
Date: 2026-04-20
Status: Draft

Summary

Implement Imu and ImuData for the OVPhysX backend, satisfying the BaseImu and BaseImuData contracts. The IMU sensor reports angular velocity and linear acceleration in the sensor's body frame. It uses numerical differentiation of velocities (same as PhysX approach).

Validation environment: No existing task uses ImuCfg directly. Write a standalone validation script that attaches an IMU to a falling/rotating rigid body and verifies output matches expected physics (gravity acceleration, angular velocity from known torque).

Guiding Principles

• Mirror PhysX Imu implementation
• Do not modify BaseImu or BaseImuData contracts
• Numerical differentiation for linear acceleration (same as PhysX — ovphysx does not provide native acceleration)

Contract to Satisfy

BaseImu (source/isaaclab/isaaclab/sensors/imu/base_imu.py)

Inherits SensorBase. Additional abstract:

Property/Method	Type	Description
data	BaseImuData	IMU data container
_initialize_impl()	method	Setup rigid body view and offsets
_update_buffers_impl(env_mask)	method	Fetch state and compute IMU outputs

BaseImuData (source/isaaclab/isaaclab/sensors/imu/base_imu_data.py)

Property	Shape	Type	Description
ang_vel_b	(N,)	wp.vec3f	Angular velocity in body frame [rad/s]
lin_acc_b	(N,)	wp.vec3f	Linear acceleration in body frame [m/s^2], includes gravity

Architecture

File Layout

source/isaaclab_ovphysx/isaaclab_ovphysx/sensors/
├── __init__.py
├── __init__.pyi
├── kernels.py           (shared sensor kernels)
└── imu/
    ├── __init__.py
    ├── __init__.pyi
    ├── imu.py           (~300 lines)
    ├── imu_data.py      (~80 lines)
    └── kernels.py       (~100 lines)

Implementation Pattern

imu.py:

1. _initialize_impl():

   • Find target prim and its rigid-body ancestor
   • Resolve fixed transform from ancestor body to target prim (done once)
   • Compose with configured sensor offset
   • Get body state from ovphysx via TensorBindings:
     • RIGID_BODY_ROOT_POSE or ARTICULATION_BODY_POSE for transforms
     • RIGID_BODY_ROOT_VELOCITY or ARTICULATION_BODY_VELOCITY for velocities
     • Body COM data for lever arm correction
   • Query world gravity from OvPhysxManager (needed for gravity bias)
   • Allocate buffers including prev_lin_vel_w for numerical differentiation

2. _update_buffers_impl(env_mask):

   • Read body transforms and velocities from ovphysx bindings
   • Read COM positions (may be CPU — copy to GPU)
   • Launch imu_update_kernel per environment:
     • Apply sensor offset quaternion to body orientation
     • Correct linear velocity for COM offset (lever arm cross product)
     • Numerically differentiate: lin_acc = (vel_current - vel_prev) / dt + gravity_bias
     • Rotate angular velocity and linear acceleration to body frame
     • Store results and update prev_lin_vel_w

imu_data.py:

• Simple container with _ang_vel_b and _lin_acc_b warp arrays
• Properties return the arrays directly

OVPhysX API Requirements

API	Purpose	Notes
Body transforms	Orientation for frame rotation	Already available via Articulation bindings
Body velocities	Linear + angular velocity	Already available
Body COM	Lever arm correction	Already available (CPU)
Gravity vector	Acceleration bias	From OvPhysxManager or USD scene

No new ovphysx APIs needed — all required data is already available through existing TensorBindings used by the Articulation. The IMU just reads body state from a different entry point (the sensor's target body rather than the root).

Warp Kernels

Kernel	Purpose
imu_update_kernel	Per-env: offset composition, COM correction, numerical diff, frame rotation
imu_reset_kernel	Zero velocity history and output buffers

These are direct ports of the PhysX kernels — the math is identical, only the data source changes.

Tests

Backend-specific tests (copy from PhysX)

Source: source/isaaclab_physx/test/sensors/test_imu.py
Target: source/isaaclab_ovphysx/test/sensors/test_imu.py

Changes: swap physics config to OvPhysxCfg.

Validation script

Write source/isaaclab_ovphysx/test/sensors/check_imu.py:

• Create a rigid body with known initial angular velocity
• Attach IMU sensor
• Step physics
• Verify ang_vel_b matches expected value
• Verify lin_acc_b includes gravity (~9.81 m/s^2 downward in body frame)

Dependencies

• Requires PR ovphysx backend integration #4852 merged
• No new ovphysx APIs needed
• Does not depend on [OVPHYSX] RigidObject asset #5316 — can use Articulation body bindings directly

Estimated Scope

• imu.py: ~300 lines
• imu_data.py: ~80 lines
• kernels.py: ~100 lines
• Tests: ~150 lines of adaptations
• Validation script: ~100 lines