Alternatives to Carla

Open-source Unreal Engine plugin that turns UE4/UE5 scenes into labeled synthetic image datasets for computer vision.

Open-source Python SDK and CLI for training and deploying physical AI models — VLA and SLM — from managed cloud compute to real robot hardware.

License the Nuro Driver for L4 autonomous driving across robotaxis, commercial fleets, and personal vehicles.

Open-source simulated benchmark for language-conditioned, long-horizon robot manipulation from the University of Freiburg.

Open-core LiDAR SLAM for ROS 2: LiDAR odometry, LIO, 3D SLAM and georeferencing you can modify at the source level.

Raw radar data — real or synthetic — feeding end-to-end autonomous driving stacks through your world model.

Open source dev kit for AI-native robotics with leader-follower teleoperation.

Build, code, and program OpenCat quadruped robot kits — Bittle, Nybble Q, and the new Quaddle desk robot dog.

OpenCOOD is an open cooperative perception benchmark for V2V 3D detection and LiDAR fusion research.

Sharbo (Forsy) is a skill network for AI agents — encode expert workflows once, then let agents activate the right skill autonomously.

BSD-licensed C++ libraries and Python bindings for SLAM, 3D vision, and sensor fusion — MRPT 3.5.2 docs.

Applied AI research on physical-world systems: transformer design, autonomous counter-UAS, and dexterous manipulation.

Modular humanoid robot platform for builders collecting data, training skills, and deploying autonomous robots.

Skild AI is building an omni-bodied robotic brain to control any robot for any physical task.

Rhoda AI builds video-predictive robot foundation models that learn heavy industrial manipulation from a single human demo.

Physical Intelligence builds vision-language-action robot foundation models that control any robot for any manipulation task.

Wayve AI Driver: mapless, vehicle-agnostic autonomy that learns to drive from data, from L2 to L4.

Odyssey-2 Max is a causal world model that simulates physical scenes in real time as you feed it actions.

6thSense captures tactile sensing and egocentric touch data for training dexterous robot policies

Cortex AI supplies real-workplace egocentric video and robot trajectory data for training embodied AI and robot foundation models.

Trim is a foundation model for physics simulation that predicts how real-world systems evolve in time, trading exactness for latency.

Remoroo is an industrial robot manipulation platform that learns dexterous tasks from a sentence and proves each skill with an auditable

Zeon Systems turns a plain-language experiment prompt into a robot-executed wet lab protocol, with a world-model preview before anything moves.

Sureform turns your company's operational data into a licensable AI training asset and pays you for it