Voyage's mission is to super-charge communities with autonomous vehicles. The company's autonomous fleet provides on-demand, door-to-door transportation for communities across the U.S., starting with The Villages communities in San Jose and Florida. When fully operational, more than 125,000 senior citizens and community residents will have the ability to summon a self-driving car to their doorstep using the Voyage mobile app, then travel anywhere within the bounds of their community.
Company Size: 100+
Industry: Self Driving Cars
Products: Sensor Fusion Cuboids
Location: Santa Clara, USA
Finding the right partner for accurate data annotation
Founded by Oliver Cameron, Warren Ouyang, Eric Gonzalez and MacCallister Higgins, all formerly of Udacity, Voyage found itself searching for high-quality, diverse datasets to use in training their autonomous algorithms.
Many open source datasets had poor annotation quality, and often lacked training data specific to Voyage's initial community deployments, like golf carts.
In deciding to collect its own datasets, Voyage knew it needed the right partner to quickly and accurately annotate their data.
For Voyage, quality training data is absolutely critical in meeting development timelines, improving perception capabilities, and ultimately delivering a world-class autonomous rider experience.
“When I think about what Scale brings to the table, it comes down to machine learning expertise. Scale has a team of very talented engineers who really understand the problems in training machine learning models, of data annotation, and of data quality.”
Scale's Sensor Fusion: combining rich tooling with highly trained human labor
Voyage looked into various vendors for data annotation services. Scale was one of the only industry options to offer both camera and LIDAR annotation, and offered the most impressive functionality in the new field of annotating LiDAR frames. With other vendors, Voyage would have needed to develop additional tooling and provide training for the human annotation labor. Scale's Sensor Fusion Cuboids product, combining rich tooling and highly trained human labor, was the perfect fit.
"Our perception team has been very happy with the results we've seen from Scale. The accuracy has been great," said Gonzalez. "Annotation is something you want to click and forget," Ouyang added, "and Scale has delivered on the quality they've promised." Scale has quickly become deeply integrated into Voyage's self-driving development stack. Leveraging Scale's rich set of APIs, Voyage has developed an internal "Send to Scale" button allowing its engineers to send data to Scale with a few simple clicks.
Additionally, on-demand support and regular communication through a private Slack channel has made Scale a true partner in achieving Voyage's mission. "When I think about what Scale brings to the table, it comes down to machine learning expertise. Scale has a team of very talented engineers who really understand the problems in training machine learning models, of data annotation, and of data quality," said Cameron.
“Data is our life blood. It's what allows us to train all of our complicated models.”
Training Data to Bet On
Voyage will continue to leverage Scale's latest features, including the new LiDAR Debug tool to ensure accurate calibration and alignment between cameras and LiDAR. Voyage also plans to increase the volume of data annotation tasks. The partnership with Scale will allow Voyage's engineering team to focus on the development of their autonomy stack, with high confidence in critical training and validation data.
“Scale has been a great partner because they have the domain expertise we're looking for. We don't need to spend time getting them up to speed on our annotation requirements.They get it. Scale has well documented APIs that support our workflows. That makes the process significantly easier to upload the data we collect from the vehicles on daily basis to their platform.”