Established in 2021, the WoRV team is a specialized Physical AI group within maum.ai, with research expertise and field experience in robotics and autonomous driving. With the goal of enabling autonomy in industrial settings, we develop AI for agriculture, logistics, defense, construction, and manufacturing using data collected from hundreds of robots performing real-world tasks.
WoRV Technology
We train AI on data from real and simulated environments and deploy it to real robots and autonomous vehicles. Data collection, simulation, training, and deployment run as one feedback loop.
Teleoperation & Data Pipeline
Collecting high-quality data for robot learning requires intuitive control across different machines and a consistent data collection process. We integrate teleoperation systems with existing hardware to record skilled operators at work, building pipelines that collect and organize the data into formats suitable for training.
Real2Sim
In the real world, limited robot fleets and space constrain data collection, and dangerous situations are difficult to attempt repeatedly. Using Unreal Engine 5 and NVIDIA Isaac Sim, we recreate the appearance, physical properties, and dynamics of real environments so robots can safely experience and learn from diverse conditions and unexpected situations.
VLA & World Models
To handle situations absent from their training data, robots need to understand their surroundings and task context before deciding what to do next. Using visual, language, and action data, we train VLA models for action selection and world models for predicting changes in the physical world, developing models that can act toward a given goal in unfamiliar environments.
Move the divider to compare the two driving scenes.
Sim2Real
Models trained in simulation can perform differently in real environments because of differences in factors such as friction and sensor noise. We apply the models to physical machines, quantify the discrepancies between simulation and reality, and feed the results back into training to improve real-world reliability.
Real World Deployment
Industrial sites present situations that were not anticipated in the lab, making it important to turn operational problems into improvements. We deploy robots and autonomous vehicles to evaluate their decisions and actions, collecting edge cases and failure data to inform the next round of training and field validation.
Join Us
At WoRV, data collection, simulation, model training, and testing on physical machines are part of one connected process. We see how our research and engineering work in the field, then bring the problems we find back into our research.
If you want to take the technology you build all the way to working robots and vehicles, explore our open positions.
View open positionsA closer look at the research and team behind WoRV.
WoRV