SimForge
SimForge provides data infrastructure for physical AI, offering scalable digital twins and simulation-ready environments for autonomous driving and robotics development. The platform connects maps, scenarios, and simulation runs in one workflow, with an open toolkit including the simforge CLI, SimForge Gym for reinforcement learning, and Helios Engine for closed-loop sensor rendering. New accounts start with $25 in free cloud credits.
- Autonomous Vehicles
- Artificial Intelligence
- Mobility & Transportation
- Developer Tools
- Automotive Technology
- Robotics
- Software Only
Funding
Founders
Founder details are not available yet.
Product
Problem
Autonomy and physical AI teams need to test their systems across vast numbers of driving scenarios, but real-world testing is unsafe, expensive, and difficult to scale. Reproducing specific road conditions and sensor outputs requires disconnected tools and manual processes, making it hard to maintain consistent context across maps, scenarios, and simulation runs.
Solution
SimForge provides a scalable digital twin and simulation platform that keeps maps, scenarios, simulation runs, and generated artifacts connected in one workflow. Teams use SimForge Studio to import or select digital twins with lane topology and signals, author scenarios with actors and behaviors, and dispatch CARLA jobs that capture synchronized RGB, depth, segmentation, and LiDAR outputs. The platform also integrates with NVIDIA Cosmos to generate photorealistic variants across weather, time-of-day, and lighting conditions without changing the underlying trajectory, and with Cosmos-Reason2 to generate structured, explainable commentary on scene understanding. SimForge SimCloud extends this to a browser-based environment where teams compose scenarios and run them on CARLA workers, capturing multi-sensor output for every run. Helios Engine renders closed-loop driving simulation with all sensors on one clock, using real places rebuilt as digital twins.
Target Audience
Primary customers are research and mobility teams shipping physical AI applications, including autonomous driving developers, robotics companies, and perception engineering teams that need scalable simulation and testing infrastructure.
Features
- Connected workflow spanning map preparation, scenario authoring, simulation execution, and artifact generation for reproducible results
- SimForge Studio desktop authoring environment with lane topology, signals, and runtime metadata management
- CARLA-based simulation dispatch with synchronized multi-sensor output including RGB, depth, segmentation, and LiDAR
- SimCloud browser-based platform for composing scenarios and running them on CARLA workers
- NVIDIA Cosmos Transfer integration for generating photorealistic render variants with identical actors and trajectories
- Cosmos-Reason2 VLM integration for structured, queryable commentary on trailing-camera footage
- Custom digital twin creation service for real-world locations with pre-authored scenarios ready to run
- Open toolkit under Apache-2.0 including the simforge CLI and Python clients, with a runtime free for non-commercial use
- SimForge Gym, a Gymnasium-compatible reinforcement learning environment with a measured, reproducible PPO baseline
- SIMLOG-25k dataset of real-world scenarios and Recast for converting real drives into editable scenarios