In our work, deeply integrated with our customer’s needs, we’ve only gotten more committed to the idea that a truly vital simulation solution should always strive to be more aligned with the real world, and always be becoming more accessible. Both are vital for our customers to meet their ambitious goals.
In our previous release (Falcon 6.1), we introduced Vibe Sim, a browser-based, agentic simulation co-pilot designed to make physically grounded digital twin simulation dramatically more accessible. By combining natural language interaction with Falcon’s high-fidelity simulation capabilities, Vibe Sim opened the door for more engineers, researchers, and AI developers to rapidly customize scenarios, generate synthetic data, and iterate on Physical AI models without the barrier of deep expertise in traditional simulation and graphics tooling workflows.
The advances arriving with Falcon 6.2 advance both prongs of this approach. This includes virtual sensor improvements that align real time simulation to even more closely match real world operations, agentic advancements that make it even easier to work with autonomous systems in simulation, and expanding Vibe Sim access to a larger segment of our users.
As always, in this blog we highlight just a portion of the features arriving with Falcon 6.2, and you can see the full release notes here.
Let’s dive into Falcon 6.2.
Major LiDAR Performance Improvements
As sensor fidelity and coverage requirements continue to increase across robotics, defense, and autonomous systems, scalable LiDAR simulation becomes increasingly critical for generating realistic synthetic datasets and validating perception stacks under operational conditions. In short, this means that our customers increasingly need the ability to generate concurrent LiDAR data streams while maintaining high frame-rates with real-time performance.
Falcon 6.2 introduces a new ray-traced LiDAR mode that samples points directly from an RT shader rather than reconstructing point clouds from image captures. The result is improved point cloud quality alongside substantial performance gains that scale with both environment complexity and sensor count.
RZR Polaris vehicle equipped with 4 simulated commercial LiDARs operating in real time in a complex, foliage rich environment (Pierson Park).
In practice, teams running complex maps with multi-LiDAR configurations can see performance improvements ranging from roughly 2x to 4x depending on scenario complexity, sensor setup, and hardware configuration. In internal benchmarking with five simultaneous LiDARs running on an NVIDIA 3080 GPU, Falcon’s RT LiDAR achieved approximately 3x faster performance compared to the previous virtual LiDAR sensor version.
These improvements are driving better scalability for multi-LiDAR configurations, making real-time simulation workflows significantly more practical for autonomy, robotics, and defense applications operating in dense and highly complex environments.
Work on the RT LiDAR sensor development also prompted a new optimization approach that delivers notable performance upgrades for the standard Capture Sensor-based LiDAR for simple and complex maps.
Both of these LiDAR modes can be utilized by our users starting today.
While these improvements are especially important for teams developing autonomy systems that rely on dense sensor stacks and large-scale environmental interaction, the new RT LiDAR mode opens the door to further LiDAR simulation development, enabling us to even better match specific LiDAR features and aligning ever closer with our customers’ use cases.
New Behaviors Agent in Vibe Sim
Falcon 6.2 also expands what Vibe Sim agents can directly control during a live session. One of the most powerful features of Falcon is that it exposes all public variables and functions for Unreal Engine objects via Python, enabling users to change variables and behaviors in real time. With this release, Vibe Sim’s Behaviors Agent can now reliably modify and execute Python scripts in-session, enabling dynamic workflows such as:
- Updating ROS2 controllers
- Modifying sensor configurations
- Changing scenario logic
- Augmenting twin behaviors
- …and much more
Two examples of behavior agent operations: (1) Adding a ROS controller to an RZR vehicle digital twin and modifying velocity, and (2) Adding a behavior that causes the RZR twin to always drive in the direction of the current location of the fixed base (Falcon’s user operated controller).
How does this work under the hood? When the Main Agent needs to modify twin behavior, it delegates it to this sub-agent, describing the desired behavior in natural language. The Behaviors Agent then produces validated Python code, which the Main Agent reviews and writes to the appropriate script file.
The addition of this capability represents an important step toward a more fully agentic simulation workflow, where users increasingly focus on defining goals, problem solving, and model iteration, while the simulation environment handles the implementation details.
Vibe Sim Expands to FalconCloud
Falcon 6.2 further deepens the integration between Vibe Sim and FalconCloud, enabling a more complete cloud-native simulation workflow.
FalconCloud already serves as the central hub for Falcon users — hosting the digital twin catalog, scenarios, datasets, and collaborative simulation workflows. With Falcon 6.2, users can now move beyond simply launching and collecting data from shared scenarios. Scenarios and twins stored in FalconCloud can now be opened directly inside Vibe Sim for live modification and iteration through new “Run Sim” and “Edit Sim” workflows.
A walkthrough of accessing Vibe Sim directly in FalconCloud, and the integrated workflow that enables rapid data generation and utilization.
This enables teams to rapidly iterate on scenarios, modify behaviors, generate synthetic data, and push updated scenarios or generated outputs back into FalconCloud — all without leaving the Falcon ecosystem.
Combined with expanded cloud compute options, users can also run Jupyter notebooks directly against generated datasets, supporting training, evaluation, and analysis workflows entirely in the cloud.
Supporting New Community and Research — Vibe Sim Opens to EDU Users
EDU users now receive up to 10 hours per month of Vibe Sim usage through FalconCloud access. This expansion reflects a broader goal behind Vibe Sim: lowering the barrier to physically grounded simulation while enabling more users to build, test, and iterate on Physical AI systems — especially for the next generation of engineers.
All EDU users can take advantage of Vibe Sim starting today, including with our latest Kaggle Challenge, which leverages Vibe Sim workflows and GIS-based digital twin environments to explore new approaches for synthetic data generation and vision model development.
This challenge, created in partnership with Lunate AI, is based on real-world geospatial AI applications where annotated data may be difficult and time-consuming to source.
Participants train a model on synthetic aerial images of various landscapes, with the ability to add relevant twins such as trees, rocks, or cars, all generated by participants in Vibe Sim. Then, users test the model’s performance on previously unseen real-world images.
As simulation becomes increasingly central to training and validating real-world AI systems, we believe broadening access to these workflows — from enterprise deployments to research and education — will play an important role in accelerating the next generation of AI, robotics, and autonomy development.
