This is a great point - vision is often all that's needed for these industrial environments.
+ We've seen that visual context from RGB data is essential for eliminating false positives. The challenge isn't distinguishing if the obstacle exists, it's distinguishing what it is so the machine can operate safely:
An autonomous excavator can run through a dirt pile, but the story changes when the obstacle is a crouched-over person. The problem with Lidar is that it looks the same on a blurry point cloud.
Bedrock Robotics@BedrockRobotics
An excavator tops out at 3 mph. A self-driving car on a highway needs to see hundreds of meters ahead at 65 mph. Our machines don't need anything close to that range, which means today's sensor technology already exceeds what we require. That margin translates directly into safety. Hear from Clayton Harper (Machine Learning Engineer), Will Grossman (Hardware Engineer), and Kenneth Hong (Hardware Systems Architect) on how the hardware ecosystem has caught up to the challenge of building for 80,000-pound machines on active job sites.
English
