bornish
lvl.3
Flight distance : 91447 ft
United Arab Emirates
Offline
|
As great as it sounds, I would be extremely skeptic on the effectiveness of the collision avoidance system of the M100. We should welcome such research from manufacturers and applaud their efforts because even small steps in the right direction are useful steps and better than no steps. My main concern remains though regarding its current implementation. Personally I would not spend $999+ something that does not come with any technical specs, but the price is not the issue. Please allow me to explain:
When designing such a safety feature for a system (UAV, car, ...) we need to primarily focus on:
- delay due to real-time complex operations which would limit the effectiveness of the feature
- health-check required to ensure that the feature is correctly calibrated and functional
- manual-override implemented, as no automatic system should be allowed to fully control a system
Basically, a collision avoidance feature should deal more with preventive flight control instead of reactive flight control. What I've seen in the animated gif was that the M100 was "repelled" by someone pushing a wooden-plate in its proximity. That could simply be achieved by a sonar reading with almost no computational requirement. It really looked to me like a wrong approach though. I would be more than happy to have a system which does not get "repelled" by objects flying into it, but instead to limit my flight decisions to prevent my system to collide into fixed objects around it. Think about it like this:
- assume the environment isn't moving and the only moving object is your UAV
- design and implement the safety feature to limit your UAV's movements based on the detected environment
- limit speed based on the proximity of detected objects, taking into consideration the measurements precision and any processing delay
- parametrize the formulas used in processing and extensively test with different configuration values (proximity allowed, max wind speed,...)
Once such a system has been successful (acceptable error margin) research focus can be shifted to the next level where more advance algorithms would try to anticipate and predict environment changes (ex. child running towards your spinning blades).
Thanks for reading.
Regards,
Bogdan |
|