Perception Team

Our Challenge

I lead the perception team at a major tech company. It's no secret that high quality data is the key to improving machine learning algorithms for object detection, depth estimation, etc. Collecting and annotating this data has been an expensive and time consuming process. We rely on third-party annotation services, but they can't keep up with our annotation needs and rely on inaccurate manual labor. Further, we need to put real cars on the road to collect this data in the first place, so it's difficult for us to capture rare cases and whenever we change sensor configurations, we have to re-collect and re-annotate the data.

The PD Solution

Parallel Domain's image generation API allows us to rapidly render highly realistic synthetic data sets with a wide variety of parameters. We get perfectly annotated data near-instantly, accelerating our data acquisition and annotation process by orders of magnitude. Further, when we need to change sensor configurations, we simply change our arguments to the API and render again. These data sets have much richer image layers / metadata than real-world labeled data sets, enabling my team to do more types of training. This was something we could not do with real-world data; we feel like PD opened up a whole new realm of machine learning for perception teams!

Auto Manufacturer

Our Challenge

I run the simulation team at one of the world’s largest car manufacturers. My company has over 250,000 employees globally and has been around for over 60 years.

I’ve been tasked with enabling end-to-end simulation of our vehicles in a wide range of realistic, real-world environments. However, my team has found environment and scenario generation to be a major bottleneck.

I’ve tried hiring computer graphic artists to do this, but the effort to generate all of the layers required for our simulation is a complete blocker. For each simulation world, we need collision meshes for our Modelon vehicle dynamics, Unreal levels for camera, surface geometry for lidar, HD map data for our ego vehicle, and semantic labeling information for our ground truth image generator. Building all of these layers by hand at the scale we need is hamstringing our simulation efforts.

The PD Solution

Without Parallel Domain’s solution, we wouldn’t be simulating at the scale or quality that we are today. Their software has jumped us ahead significantly on our simulation timeline by alleviating the content generation bottleneck. This allows us to focus precious resources on hitting our autonomy milestones.

We’re now able to generate 3D environments directly from our map data as part of our continuous integration pipeline. The generated result includes all of the necessary layers for our full stack simulation. These layers integrate well with our simulation components since PD outputs to various industry standard formats. From our lidar simulator to our dynamic agents framework, we now have the content we need to focus our team’s energy on the most important thing: making our cars drive themselves.

Autonomous Vehicle Company

Our Challenge

I am the Director of Simulation at a leading autonomous vehicle startup. We’re reinventing transportation from the ground-up, and that includes building our own core simulator. However, we need to fill that simulator with a massive variety of scenarios and environments, something that we cannot take the time to build in-house.

We’re about to start running full-stack simulation at scale and our outsourcing vendor is manually building these environments one at a time. If we don’t find a more automated solution, our simulation roadmap would be delayed significantly.

The PD Solution

I have been working with Parallel Domain for over a year now and they have been stellar. Not only is the team a pleasure to work with, but their technology is leaps beyond what we would have been able to develop internally. I’ve been able to quickly generate different environments and experiment with variables such as road topology, number of lanes, elevation variations, types of vehicles, and even dynamic traffic behavior.

Even in the absence of input HD map data, PD’s solution outputs an HD map that perfectly matches the generated 3D world. This allows me to run our ego vehicle simulation while interacting with PD’s dynamic agents in environments that range from small functional tests up to entire city neighborhoods.

Semiconductor Manufacturer

Our Challenge

I am the Head of Sales at a leading GPU manufacturer. We serve customers from various industries but now we have a specific focus on growing our business in the AV industry, especially simulation.

We are experts in semiconductors, so we’re developing chipsets with teraflops of computing power that can run ray-traced sensor simulation in real time. While we have experience building 3D content by hand, we have been unsuccessful in generating enough high fidelity environments for both our internal teams’ and external customers’ simulation needs. We need an automated solution for producing large scale environments at various levels of detail.

The PD Solution

Our partnership with Parallel Domain has been integral to accelerating both our simulation roadmap and the development of our hardware in the loop simulation ecosystem.

We can use their software to seamlessly generate virtual environments that reside within our own simulation platform that we deliver to both our internal teams and external customers. Their technology can generate environments across a large level of detail spectrum. This allows us to push our hardware to the limit, stressing graphical throughput on the high end, and pushing high framerate on the symbolic end. Whether we’re shooting for cutting-edge sensor simulation or scenario simulation at hyper fast frame rates, PD’s solution has us covered.

Tier 1 Simulation provider

Our Challenge

I lead partnerships at a large simulation software company. We supply a simulation-as-a-service platform to OEMs, sensors manufacturers, machine learning researchers, and more.

We want to minimize the barrier to entry for our customers and our goal is to 1000x their daily simulated miles, but this requires an enormous variety of environments and scenarios and we don’t have the time or capacity to build content generators internally.

We did not have a scalable way to offer different environments as a part of our software package without manually building maps and environments. This problem was hampering our customers’ ability to simulate large numbers of miles, slowing the adoption of our platform. We’re actively overbooked with customers asking for new 3D environments.

The PD Solution

With access to PD’s massive content library, our customers are multiplying the number of miles that they’re simulating each week. At the end of the day, this means that we’re driving more revenue through our platform, both supercharging our service as well as enabling our customers to more rapidly train and test their autonomous systems.

Further, when customers have bespoke content needs, we’re able to use PD’s software to generate and re-generate different environments, no longer throttled by waiting for manually constructed environments.

We’re looking forward to a long and fruitful partnership with Parallel Domain as we scale up our platform.

See ways we could partner to generate virtual environments and accelerate time to safety? Want to find out more about our services?