Hey everyone, I've been working on robotics projects for a while now. As sensor prices drop, there's a growing trend to use more sensors in projects. But do you think increasing the number of sensors is more important, or is developing smarter algorithms the way to go? For example, would it be more efficient for a simple robot to perceive its surroundings using 2 sensors combined with a deep learning-based algorithm instead of 10 sensors? Which approach do you prefer?
In robotic projects, is sensor trickery more important or algorithmic intelligence?
👁️ 4 views💬 2 replies❤️ 0 likes
2 Replies
This really seems like a tough question to answer, but in reality, the balance between the two varies from project to project. During my time working on the eighth-generation iPhone, I repeatedly saw the need to minimize the number of touchscreen sensors while optimizing the algorithm—because every extra sensor would multiply power consumption and cost. The same principle applies in robotics; for example, in autonomous cars, using deep learning with cameras instead of expensive sensors like LIDAR (Tesla’s approach) reduces both cost and computational load.
However, there’s a key point to consider here: sensors don’t just provide input data for the algorithm—they also determine error tolerance and reliability. For instance, a robot using just two cameras and a deep learning model might have a higher detection error than a system with 10 sensors, because the algorithm struggles to compensate for blind spots caused by the camera’s field of view. The best strategy here is to minimize the number of sensors based on the system’s requirements and fill the remaining gap with algorithmic improvements.
More and more companies are trying to strike this balance; for example, NVIDIA’s Jetson platform or Google’s Coral products efficiently bridge the gap between sensor optimization and algorithmic computing power. So, the answer depends on the project’s goal—if cost and power consumption are critical, reducing sensors and improving the algorithm makes sense; if maximum reliability and precision are required, increasing sensors and simplifying the algorithm might be the smarter choice.
You've made a valid point, folks. Based on my experience, using high-quality algorithms is much more efficient than simply increasing the number of sensors. In my early projects, I also operated under the mindset of "the more sensors, the better," but later realized that data quality and algorithm improvements are far more critical than the number of sensors.
Last year, in an autonomous equipment robot I built, I implemented a deep learning-based system to distinguish paths using just three sensors (a LIDAR, an RGB camera, and an IMU). Previously, with ten budget-friendly sensors, the system was completely unstable, but now it's both cheaper and more reliable. Minimizing sensors while optimizing the algorithm makes much more sense in terms of both budget and performance.