Simulation platform for testing, evaluating, analyzing, and training AI models at scale. Ensuring public safety while accelerating autonomy development.
Receipts attached.
Every mile your fleet captures is a test you haven't run yet.
PD Replica turns raw capture logs into verifiable closed-loop simulation environments. Geometry. Physics. Segmentation. Matched lighting. HD maps. Calibrated camera, lidar, and radar.
All shipped with a sim-to-real report that quantifies how close to reality each PD Replica is.
Production-grade reconstruction. Verifiable simulation at scale. That's the stack leading autonomy programs actually run on.
What's the toughest mile your fleet encountered that you can't reproduce in sim?
#PhysicalAI #AutonomousVehicles #Simulation #SensorSimulation #Reconstruction #Replica
Building Physical AI that works in the real world requires rigorous testing at scale. Today we're sharing how we work with @NVIDIA to deliver production-grade simulation to Physical AI developers across automotive, robotics, drone delivery, and more.
Post here - https://t.co/CKdbyoojAy or come see a live demo at #NVIDIA #GTC2026 booth 1647
🚀 Parallel Domain @ @NVIDIAGTC 2026
We’re excited to announce that Parallel Domain will be exhibiting at NVIDIA GTC 2026. Come see how we're powering the next generation of physical AI with high-fidelity sensor simulation and scene reconstruction.
📍 Booth 1647
📍 San Jose Convention Center
📅 March 16–19, 2026
Stop by our booth or reach out to schedule a private demo with our on-site crew.
See you in San Jose!
#NVIDIAGTC #PhysicalAI #Robotics #AV #Simulation #EmbodiedAI #Evaluation #Testing #SoftwareDefinedVehicles #ADAS #Autonomy
Two years ago, you had to choose between realistic or scalable simulation.
Today, that tradeoff is fading.
Software can reconstruct the environment and dynamic actors from real-world logs, bring them into simulation, generate variations, test perception systems, and validate model fixes at scale.
This is how autonomy teams move faster without relying on endless real-world miles.
At Parallel Domain, we're seeing the data flywheel go from theory to practice, and it's changing how safety and deployment happen at scale.
Thank you Evan Helda for the great conversation with Kevin McNamara at the @nebiusai #PhysicalAI and #Robotics awards.
"If your code took days to weeks to compile and test, it would be nearly impossible to develop good software."
Unfortunately some autonomous vehicle testing is still stuck in that world, where the "compile and test" step is real-world testing, track time, safety drivers, and limited scenario coverage.
In our recent blog post we argue why autonomous vehicle testing needs a shift left using data-driven simulation and digital twins.
Read it here: https://t.co/e7BoaNrqtV
#AutonomousVehicles #Safety #Simulation #DigitalTwins #SoftwareDefinedVehicles
New at #CES: We’re integrating @NVIDIA Omniverse NuRec Fixer into our PD Replica pipeline!
Why this matters for physical AI: simulation needs digital twins that are geometrically accurate and resilient to real-world capture gaps (occlusions, missing viewpoints). Fixer helps reduce artifacts and improve novel-view rendering, turning messy sensor data into cleaner, simulation-ready environments.
We are proud to be collaborating closely with NVIDIA to empower physical AI developments.
Learn more by stopping by our booth at CES LVCC West Hall 6673 and read more here - https://t.co/X9Beu66aM1
#SoftwareDefinedVehicles #Simulation #PhysicalAI #Autonomy
Last week, at the Robotics & Physical AI Awards, the Training Ground panel focused on how robotics teams rely on data engines, simulation and large-scale compute long before a robot enters the real world.
@foxglove, @parallel_domain and @anyscalecompute leaders discussed the infra and workflows that make scalable autonomy possible.
#PhysicalAI #Robotics #AIInnovation
@nebiusai@PalatialSim@dualityai Thank you for the public recognition of our technical achievements @nebiusai! We spend our days on the long road to building more realistic, deterministic simulation… so of course we had to celebrate with an extra-long drive in sim!
Here’s to many more miles together. 🙌
Announcing the 2025 Robotics and Physical AI Awards winners:
Data Engines and Simulation
1st Buildroid
2nd @parallel_domain
3rd @PalatialSim and @dualityai
(Thread 1/5):
@nebiusai@PalatialSim@dualityai Thank you for the public recognition of our technical achievements @nebiusai! We spend our days on the long road to building more realistic, deterministic simulation… so of course we had to celebrate with an extra-long drive in sim!
Here’s to many more miles together. 🙌
Every kilometer your fleet drives can become a new multi-kilometer scale test track.
We digitally reconstruct the world captured by your sensors, enabling real-fidelity, deterministic simulation.
From those drive logs, we generate simulation‑ready digital twins with simulated ground‑truth across camera, lidar, near-infrared, and radar so developers can hammer on the same locations with near‑infinite, deterministic variation.
In the video below you’ll see us:
- Turn camera logs from 4 real locations into kilometer‑scale PD Replicas.
- Simulate lidar, radar and multi-camera sensor rigs.
- Auto‑generate annotations (RGB, 2d/3d bounding boxes, depth, segmentation).
- Re‑simulate the exact same stretch of road with deterministic variation.
Drive once. Re‑test forever.
Expand coverage without hand‑authoring maps or building new physical tracks.
#autonomy #simulation #digitaltwin #lidar #radar #perception
We’re excited to announce our partnership with @ForetellixHQ! Together, we’re delivering a next-generation simulation solution that fuses Foretellix’s scenario-based validation with Parallel Domain’s photorealistic sensor simulation and digital twins. Enabling AV teams to test and validate AI models with unprecedented realism, control, and scalability.
This integration helps developers test smarter, faster, and with measurable safety coverage that’s impossible to achieve with real-world testing alone.
Learn more about how we’re advancing safe autonomy here - https://t.co/oZ7TWWt82L
#AutonomousVehicles #Simulation #AI #DigitalTwin #AVTesting #SafetyValidation #ParallelDomain #Foretellix
Veronika Nihlén explores how end-to-end driving and neural reconstruction are reshaping simulation, and why it matters for safety.
Other topics in this episode include:
• Why simulation is the safest place for failure “It takes a billion miles to make a safe AV… and one bad mile to ruin it.”
• From procedural generation to reconstructive simulation, and how NeRFs and Gaussian splats changed everything
• The “bitter lesson” of AI progress: why data and compute beat hand-coded logic
• How PD Replica turns 10 seconds of real driving into 1,000 hours of testing
• Why simulation will underpin not just cars, but all embodied AI, from drones to tractors
Big thanks to @zenseact , Veronika Nihlén, and Erik Rosén for the great discussion with @kev_mcnamara from pixels to torque!
“Why unlearning what we have known to be true is existential." A powerful opening to an excellent new episode of the Deeper Learning podcast.
Here is a short excerpt, but 🎧 listen to the full episode here:
Spotify: https://t.co/LmUySzINPq
Apple: https://t.co/Yb9ShyLw86
#AutonomousDriving #Simulation #AI #DeeperLearning #Simulation #Replica #SoftwareDefinedVehicles #ComputerVision #AV #ADAS
Safer autonomy will not come from more miles, but from smarter testing.
Shifting left into simulation unlocks scale, speed, and safety.
👉 https://t.co/SHksHPLPPQ
#ADAS#Simulation#FutureOfMobility
Big milestone for Parallel Domain + Mcity @UMichMcity - We’ve completed a photorealistic Replica of the Mcity Test Facility at the University of Michigan.
Why it matters:
- Pixel-level realism for testing perception, planning & control
- Built from Mcity’s own drive-log data
- Enables researchers to validate in simulation and compare directly against on-track runs
The Mcity PD Replica makes it possible to test the entire AV stack anytime, anywhere. Accelerating safe autonomy with realism that scales.
👉 Learn more: https://t.co/E6dgPpIFBv
#Simulation #Autonomy #DigitalTwin #Mcity #AI #SoftwareDefinedVehicles
Pedestrian detection models often struggle with the rare, high-risk scenarios that matter most. Yet those are exactly the moments that are hardest and most dangerous to capture in the real world.
That’s why we asked: Can simulation reliably predict real-world pedestrian detection accuracy?
The answer? Across multiple models, PD Replica Sim showed just a 2–10% sim-to-real gap.
This is a critical step forward toward using simulation not just for development, but as a trusted platform for validating life-saving systems.
Full research article in thread
#softwaredefinedvehicles #ADAS #ComputerVision #Autonomy #Safety