Months of data engineering skipped with a quick install.
Fruit pickers spend nearly 30% of their time hauling material back-and-forth to a farm's warehouse over rough terrain. This prevents farms from harvesting all of their fruit each season, resulting in upwards of 20% left to rot. Farms unfortunately operate on such thin margins that hiring more workers is not an option, and the problem is only getting worse as labor becomes less available.
Fox Robotics is solving this problem with an autonomous cart which will transport fruit and other materials (e.g. soil) across the farm to free up the workers' time to focus on picking.
Engineering an autonomous robot is no easy task, so we need to collect large quantites of data at our customers' farms to analyze performance, debug failures, and build machine learning datasets. It's critical that this data comes from the real world to ensure the robot is able to operate outside of contrived or simulated settings.
However, two major roadblocks have sat in the way of effective data collection. The first is the lack of a quality connection over cellular networks. Prior to using SensorSurf, we would write sensor data to disk and offload it after transporting the entire machine to Wi-Fi. SensorSurf has enabled us to offload data over the worst connections, where latency is high and bandwidth drops to less than ten kilobits per second.
The second largest obstacle was the scale of data. Our robot has cameras and other sensors to perceive its surroundings, which produce multiple gigabytes of data per minute. This fills up the disk rapidly with mostly useless data. SensorSurf has enabled our fleet to record data on events, conserving space and exposing only the most interesting moments.
Once our data makes it to the cloud, we've been empowered to put the data to use. If there was a failure such as an emergency stop, we're able to immediately stream data directly into Foxglove Studio and digest what happened. Metrics are tracked with ease using SensorSurf's Grafana integration. And machine learning datasets are easier-than-ever to build using natural language search over imagery.
Our team is incredibly excited to work with SensorSurf. If you face any of the above challenges, don't hesitate to get started with their platform today.