Simulation infrastructure for physical AI: synthetic data, robot policy evaluation, and benchmarks.https://www.bifrost.ai/https://www.bifrost.ai/blog/pi05-libero-results/https://www.bifrost.ai/blog/pi05-libero-results/We ran the public pi0.5 LIBERO checkpoint for 1,999 episodes across all four suites and averaged 96.7%, against 96.85% reported by openpi. Per-suite results and rerun spread.Tue, 06 Oct 2026 00:00:00 GMThttps://www.bifrost.ai/blog/robot-policy-evaluation-platforms-compared/https://www.bifrost.ai/blog/robot-policy-evaluation-platforms-compared/Isaac Lab-Arena, RoboLab, LeRobot and SimplerEnv evaluate robot policies in simulation, RoboArena and AutoEval on real robots. The right one depends on your benchmark.Tue, 06 Oct 2026 00:00:00 GMThttps://www.bifrost.ai/blog/synthetic-data-for-defense-isr/https://www.bifrost.ai/blog/synthetic-data-for-defense-isr/Synthetic data gives ISR perception teams labeled EO/IR and SAR imagery of rare targets they cannot collect or share. Train on it with some real data, then test on real.Tue, 06 Oct 2026 00:00:00 GMThttps://www.bifrost.ai/blog/synthetic-data-for-usv-collision-avoidance/https://www.bifrost.ai/blog/synthetic-data-for-usv-collision-avoidance/Synthetic data gives USV lookout models the rare, dangerous cases real footage lacks, like swimmers, debris, night and rough seas. Train on it, validate on real video.Tue, 06 Oct 2026 00:00:00 GMThttps://www.bifrost.ai/blog/why-vla-policies-fail/https://www.bifrost.ai/blog/why-vla-policies-fail/VLA policies fail by grabbing the wrong object, missing grasps, closing on air, stalling and not recovering, and they get worse when cameras, layouts or wording change.Tue, 06 Oct 2026 00:00:00 GMThttps://www.bifrost.ai/blog/best-simulation-benchmarks-for-vla-models/https://www.bifrost.ai/blog/best-simulation-benchmarks-for-vla-models/The best VLA simulation benchmarks in 2026 are LIBERO-Plus or LIBERO-PRO, RoboCasa365, RoboTwin 2.0, RoboLab and SimplerEnv. Plain LIBERO is saturated at 97.1%.Fri, 02 Oct 2026 00:00:00 GMThttps://www.bifrost.ai/blog/does-synthetic-data-work-for-object-detection/https://www.bifrost.ai/blog/does-synthetic-data-work-for-object-detection/Yes, mostly as a complement. Pretraining on synthetic images and fine-tuning on a small real set beats real-only training in most published results.Fri, 02 Oct 2026 00:00:00 GMThttps://www.bifrost.ai/blog/how-many-trials-to-evaluate-a-robot-policy/https://www.bifrost.ai/blog/how-many-trials-to-evaluate-a-robot-policy/About 100 trials gives a 95% interval of roughly ±8 to ±10 points on one success rate. Telling two policies 10 points apart takes 199 to 388 trials each.Fri, 02 Oct 2026 00:00:00 GMThttps://www.bifrost.ai/blog/how-to-close-the-sim-to-real-gap-for-perception/https://www.bifrost.ai/blog/how-to-close-the-sim-to-real-gap-for-perception/Measure the gap on held-out real data, model your sensor, randomize appearance, match the real content distribution, then fine-tune on a small real set.Fri, 02 Oct 2026 00:00:00 GMThttps://www.bifrost.ai/blog/how-to-run-libero-evaluation/https://www.bifrost.ai/blog/how-to-run-libero-evaluation/Run each LIBERO suite from its 50 fixed initial states, 50 episodes per task, with step limits of 220 to 520, then match the training image rotation and action chunking.Fri, 02 Oct 2026 00:00:00 GMThttps://www.bifrost.ai/blog/is-libero-saturated/https://www.bifrost.ai/blog/is-libero-saturated/Largely, yes. Top VLAs report 96 to 98% averages, but LIBERO-PRO and LIBERO-Plus perturbations drop several to near 0% when objects move, tasks change or cameras shift.Fri, 02 Oct 2026 00:00:00 GMThttps://www.bifrost.ai/blog/isaac-lab-vs-mujoco-vs-maniskill-vs-genesis/https://www.bifrost.ai/blog/isaac-lab-vs-mujoco-vs-maniskill-vs-genesis/For policy evaluation, the benchmark usually picks the simulator. LIBERO and RoboCasa run on MuJoCo, SimplerEnv on ManiSkill, Isaac Lab-Arena tasks on Isaac Lab.Fri, 02 Oct 2026 00:00:00 GMThttps://www.bifrost.ai/blog/parallel-gpu-robot-policy-evaluation/https://www.bifrost.ai/blog/parallel-gpu-robot-policy-evaluation/Eval time goes to policy inference, sim stepping, rendering and resets. Vectorize environments, shard across GPUs, batch inference and serve the policy separately.Fri, 02 Oct 2026 00:00:00 GMThttps://www.bifrost.ai/blog/sim-to-real-correlation-robot-policy-evaluation/https://www.bifrost.ai/blog/sim-to-real-correlation-robot-policy-evaluation/For ranking policies, often yes. SIMPLER reports a 0.924 Pearson correlation with real results and PolaRiS 0.90, but sim success rates rarely match real ones.Fri, 02 Oct 2026 00:00:00 GMThttps://www.bifrost.ai/blog/simplerenv-alternatives/https://www.bifrost.ai/blog/simplerenv-alternatives/The main SimplerEnv alternatives are PolaRiS for DROID policies, RoboLab in Isaac Lab-Arena, Gaussian-splat soft-body twins, RobotArena ∞, and real-world RoboArena.Fri, 02 Oct 2026 00:00:00 GMThttps://www.bifrost.ai/blog/synthetic-data-for-drone-detection/https://www.bifrost.ai/blog/synthetic-data-for-drone-detection/Synthetic data improves drone detectors most when mixed with real images. Published hybrids beat real-only training, while synthetic-only models trail on hard test sets.Fri, 02 Oct 2026 00:00:00 GMThttps://www.bifrost.ai/blog/synthetic-data-for-lunar-landing-hazard-detection/https://www.bifrost.ai/blog/synthetic-data-for-lunar-landing-hazard-detection/Lunar hazard detection models train on synthetic terrain because labeled lander-scale imagery barely exists. Teams build it from LOLA DEMs, procedural rocks and low sun.Fri, 02 Oct 2026 00:00:00 GMThttps://www.bifrost.ai/blog/synthetic-data-platforms-compared/https://www.bifrost.ai/blog/synthetic-data-platforms-compared/The best synthetic data platform depends on your domain and sensors. We compare NVIDIA, Duality, Parallel Domain, Rendered.ai, Anyverse and others as of October 2026.Fri, 02 Oct 2026 00:00:00 GMThttps://www.bifrost.ai/blog/synthetic-thermal-infrared-training-data/https://www.bifrost.ai/blog/synthetic-thermal-infrared-training-data/Labeled thermal IR training data comes from three routes: hand-labeling real frames, translating RGB images to thermal, or rendering scenes with physics-based simulation.Fri, 02 Oct 2026 00:00:00 GMThttps://www.bifrost.ai/blog/your-robot-may-be-learning-the-world-flipped/https://www.bifrost.ai/blog/your-robot-may-be-learning-the-world-flipped/We found that the most widely used LIBERO training datasets store images that are horizontally mirrored relative to the simulated scene. Here is the history, the impact, and how Manifold catches this class of bug.Mon, 14 Sep 2026 00:00:00 GMThttps://www.bifrost.ai/blog/introducing-manifold/https://www.bifrost.ai/blog/introducing-manifold/The platform for accelerating robotics research with failure analysis that explains itself. A month of experiments in a week.Fri, 21 Aug 2026 00:00:00 GMThttps://www.bifrost.ai/blog/multi-camera/https://www.bifrost.ai/blog/multi-camera/Place up to 5 cameras in a single world and capture every angle of the same moment in one render.Tue, 21 Jul 2026 00:00:00 GMThttps://www.bifrost.ai/blog/how-to-evaluate-vla-policy/https://www.bifrost.ai/blog/how-to-evaluate-vla-policy/A practical guide to VLA policy evaluation, covering LIBERO and RoboCasa, how many rollouts you actually need, and a reproducibility checklist for citable results.Fri, 17 Jul 2026 00:00:00 GMThttps://www.bifrost.ai/blog/top-datasets-robot-manipulation/https://www.bifrost.ai/blog/top-datasets-robot-manipulation/A framework for evaluating manipulation datasets, plus 12 of the best open-source datasets for training VLA models and manipulation policies.Fri, 17 Jul 2026 00:00:00 GMThttps://www.bifrost.ai/blog/notebooks-live-instances/https://www.bifrost.ai/blog/notebooks-live-instances/Edit notebooks and dispatch render jobs without a GPU instance, pre-warm instances for when you need them, and share packaged datasets across your organization.Tue, 14 Jul 2026 00:00:00 GMThttps://www.bifrost.ai/blog/3d-engine-upgrade/https://www.bifrost.ai/blog/3d-engine-upgrade/Our upgraded 3D engine unlocks progressive improvements in visual quality and render performance, starting with a more realistic weather system.Wed, 08 Jul 2026 00:00:00 GMThttps://www.bifrost.ai/blog/top-datasets-maritime-ai-perception/https://www.bifrost.ai/blog/top-datasets-maritime-ai-perception/A framework for evaluating maritime datasets, plus 14 of the best open-source datasets for USV perception, obstacle detection, and ship classification, scored on ontology, diversity, and label quality.Mon, 06 Jul 2026 00:00:00 GMThttps://www.bifrost.ai/blog/docs-ai-assistant/https://www.bifrost.ai/blog/docs-ai-assistant/Get instant answers about the Bifrost platform with the new AI assistant built into our docs.Tue, 09 Jun 2026 00:00:00 GMThttps://www.bifrost.ai/blog/thermal-with-depth/https://www.bifrost.ai/blog/thermal-with-depth/Generate thermal imagery with pixel-aligned depth maps from a single Bifrost render.Tue, 19 May 2026 00:00:00 GMThttps://www.bifrost.ai/blog/free-notebook-sessions/https://www.bifrost.ai/blog/free-notebook-sessions/Explore and experiment freely in Bifrost notebooks - your credits now go entirely to rendering.Fri, 15 May 2026 00:00:00 GMThttps://www.bifrost.ai/blog/bootstrapping-maritime-detection-with-synthetic-data-in-2-days/https://www.bifrost.ai/blog/bootstrapping-maritime-detection-with-synthetic-data-in-2-days/One engineer trained a USV sailboat-and-buoy detector to 82% F1 on real test footage in two days—using just 2,500 synthetic images and two training runs on a single desktop GPU.Wed, 29 Apr 2026 00:00:00 GMThttps://www.bifrost.ai/blog/custom-asset-requests/https://www.bifrost.ai/blog/custom-asset-requests/Request bespoke 3D assets directly in the Bifrost platform and track every request from submission to delivered asset.Fri, 24 Apr 2026 00:00:00 GMThttps://www.bifrost.ai/blog/neural-rendering-preview/https://www.bifrost.ai/blog/neural-rendering-preview/Neural rendering adds photoreal detail and new environmental diversity on top of Bifrost's 3D engine, without breaking your labels.Wed, 08 Apr 2026 00:00:00 GMThttps://www.bifrost.ai/blog/panoptic-segmentation-masks/https://www.bifrost.ai/blog/panoptic-segmentation-masks/Pixel-accurate panoptic segmentation masks replace COCO polygons in Bifrost dataset downloads, now delivered as high-bit-depth TIFF files.Mon, 23 Mar 2026 00:00:00 GMThttps://www.bifrost.ai/blog/org-wide-sharing/https://www.bifrost.ai/blog/org-wide-sharing/Share notebooks, ontologies, and generations across your organization so every team builds on each other's work.Tue, 17 Mar 2026 00:00:00 GMThttps://www.bifrost.ai/blog/glassy-reflections/https://www.bifrost.ai/blog/glassy-reflections/A new ultra-calm water material brings mirror-like ocean reflections to your maritime synthetic data.Wed, 11 Mar 2026 00:00:00 GMThttps://www.bifrost.ai/blog/offline-render-jobs/https://www.bifrost.ai/blog/offline-render-jobs/Send offline render jobs instantly, name collections on the way out, and track every job with clear statuses.Wed, 25 Feb 2026 00:00:00 GMThttps://www.bifrost.ai/blog/ntt-data-november-2025/https://www.bifrost.ai/blog/ntt-data-november-2025/Over the past year, NTT DATA's Innovation Center collaborated with Bifrost AI to evaluate whether synthetic data could accelerate AI model development while maintaining or improving quality and reducing costs.Wed, 19 Nov 2025 00:00:00 GMThttps://www.bifrost.ai/blog/dataset-gallery/https://www.bifrost.ai/blog/dataset-gallery/Get instant access to ready-to-run datasets inspired by some of the world's top AI teams.Wed, 16 Jul 2025 00:00:00 GMThttps://www.bifrost.ai/blog/series-a/https://www.bifrost.ai/blog/series-a/AI's biggest frontier is the physical world. Why data is the bottleneck, and how Bifrost builds the generative data engines to simulate it in minutes.Wed, 30 Oct 2024 00:00:00 GMT