Posts: 1,149 Threads: 50 Likes Received: 704 in 450 posts Likes Given: 1,189 Joined: Sep 2016 Reputation: 30 22-Dec-2017, 10:03 AM (This post was last modified: 22-Dec-2017, 10:16 AM by fftunes.) About Kalman filter, i just read this part on rcg: "It's exactly the same implementation, just rewritten to look differently. It's unsuited for the job it's supposed to do. The Kalman gain is supposed to adapt according to the error in prediction, yet in no place is any prediction error being considered. It's a plain lowpass the way it works currently." True? I.e. is this just a question of definition, and does not using prediction error make it any less worth using? • Posts: 2,286 Threads: 38 Likes Received: 1,527 in 995 posts Likes Given: 1,881 Joined: Apr 2016 Reputation: 72 (22-Dec-2017, 10:03 AM)fftunes Wrote: About Kalman filter, i just read this part on rcg: "It's exactly the same implementation, just rewritten to look differently. It's unsuited for the job it's supposed to do. The Kalman gain is supposed to adapt according to the error in prediction, yet in no place is any prediction error being considered. It's a plain lowpass the way it works currently." True? Yes, the implementation of what's in RF1 and what RS2K has in his branch of Betaflight is essentially the same thing, opening Betaflight up to trouble if they merge it. Compare the functions InitPaf and PafUpdate in RF1's flight_controller/src/filter.c with the fastKalmanInit and fastKalmanUpdate functions in src/main/common/filter.c in RS2K's fork of betaflight in the 3.2_kalman_test branch. It's hard to see what is used and what isn't used in RF1 as there's so much dead code that's never called and lots of commented out code that makes it hard to read. You almost get the impression that they don't know how to use a version control system. Glowtape's observation that the error in prediction is not fed back into the filter is also correct. So, the filter, as implemented is not actually a proper Kalman filter. • Posts: 2,286 Threads: 38 Likes Received: 1,527 in 995 posts Likes Given: 1,881 Joined: Apr 2016 Reputation: 72 22-Dec-2017, 11:20 AM (This post was last modified: 22-Dec-2017, 11:20 AM by unseen.) (22-Dec-2017, 10:03 AM)fftunes Wrote: I.e. is this just a question of definition, and does not using prediction error make it any less worth using? That's an interesting question. In the end, flight performance is what matters. However, the criticism being levelled at both RF and Betaflight that using 32KHz sampling and super fast loop times to the detriment of filtering accuracy and correct algorithm implementation due to time constraints is probably quite valid. The KISS firmware and firmwares like dRonin and Arducopter use much slower gyro sample rates and much slower loop times so that the filters can be much smarter and use techniques like sensor fusion to integrate all the available sensor data into a much richer and more accurate model of the system's state. Does any of this matter if your quad flies well? Probably not. Posts: 2,286 Threads: 38 Likes Received: 1,527 in 995 posts Likes Given: 1,881 Joined: Apr 2016 Reputation: 72 It looks like rs2k just took the gloves off! https://ornerydevelopment.wordpress.com/ The Cease and Desist posted on the web site demands that Preston stops using RaceFlight and that he pays rs2k $250,000 to avoid court action. I have a feeling this drama is going to keep on giving through 2018. Posts: 2,286 Threads: 38 Likes Received: 1,527 in 995 posts Likes Given: 1,881 Joined: Apr 2016 Reputation: 72 I also noticed that the RaceFlight-One repo on rs2k's github account is no longer blocked via Preston's DMCA complaint to github, possibly implying that rs2k could prove that he is the copyright owner to get it unblocked. Posts: 634 Threads: 56 Likes Received: 335 in 196 posts Likes Given: 183 Joined: Mar 2016 Reputation: 15 i'm really enjoyed the drama, yet jokes aside this might be also a really tough time for the guys, looking at the code... i'd say it is good nothing to whine about so the guy actually knows his ways around programming. i look at it like this ... i've builded something or done significant contribution... and afterwards i would be left with nothing... cmon... anyone who has created something has (or should have) some pride in building such quality (comparing to BF). as for some parts being "recoded to avoid duplication" that is normal stuff... look at stack overflow - 2/3 of current programmers would be jobless without it All the best Grzesiek (Grisha/ Greg) Curently flyable: Nox 5, Minimalist 112 Bench / in progres: fixing Nox 3, Scrap thinking about building: 450 • Posts: 2,286 Threads: 38 Likes Received: 1,527 in 995 posts Likes Given: 1,881 Joined: Apr 2016 Reputation: 72 I have no idea what went on behind the scenes to cause such a rift between Kalyn and Preston but it looks like the situation is going to get lots worse before it gets better. To be frank, I have no sympathy for Preston at all. When he was caught red handed using open source code in his closed source product, the video that came out where Preston attempted to re-write history and wriggle out of any accusation of wrongdoing was enough to convince me that Preston is not a nice person. The way that he spoke over and for Kalyn, despite the fact he was sitting right next to him was painful to watch. What I don't understand is why Kalyn went along with closing the branch of Betaflight they started off with in the first place. He must have known that what they were doing was wrong. The fact that Preston has made a point of insulting Boris and the rest of the Betaflight developers, both publicly and in snide comments in the Raceflight code, speaks volumes about the contempt with which he can treat other people. Maybe he had some kind of hold over Kalyn at the time. Anyone who tries to run a business where part of the business strategy is trying to sow FUD about other projects and create feeling against them, has the wrong way of looking at the world. If you behave like Preston has done, sooner or later, karma will catch up with you. Posts: 43 Threads: 4 Likes Received: 14 in 9 posts Likes Given: 15 Joined: Nov 2017 Reputation: 1 Does per work mean that he could end up being fined 150k for each piece sold or produced? 250k is a bargain if that's the way it is. • Posts: 2,286 Threads: 38 Likes Received: 1,527 in 995 posts Likes Given: 1,881 Joined: Apr 2016 Reputation: 72 (08-Jan-2018, 02:41 AM)JukeJoint Wrote: Does per work mean that he could end up being fined 150k for each piece sold or produced? 250k is a bargain if that's the way it is. I believe it does. • Posts: 2,410 Threads: 136 Likes Received: 1,790 in 1,052 posts Likes Given: 3,302 Joined: Jan 2017 Reputation: 50 I wonder when they will make the movie! I mean the documentary has already been made for the similar happenings in the 3D printer world! (on Netflix even lol) JB and I share a very similar opinion. Posts: 1,504 Threads: 83 Likes Received: 944 in 654 posts Likes Given: 2,142 Joined: Sep 2016 Reputation: 24 Love you too JB!! Windless fields and smokeless builds • Posts: 12,099 Threads: 125 Likes Received: 3,739 in 2,836 posts Likes Given: 99 Joined: Feb 2017 Reputation: 388 Have any one of you keeping g up with the Kalman filter thread on RCG? It is get very heated over there. • Posts: 1,504 Threads: 83 Likes Received: 944 in 654 posts Likes Given: 2,142 Joined: Sep 2016 Reputation: 24 I don't really visit RCG much anymore, it reminds me too much of London, too many people in too much of a hurry. As JB says it's all about sharing, if people didn't share their knowledge and experience of the hobby, there would be far fewer people enjoying it. Those people bring money to Dev's to enhance the hardware, and again sharing ideas of how to move forward to improve software to take better advantage of enhanced hardware. As the 'tagline' of the forum used to be (and still is on Oscars blog) "Share Knowledge and Ideas" Windless fields and smokeless builds Posts: 2,410 Threads: 136 Likes Received: 1,790 in 1,052 posts Likes Given: 3,302 Joined: Jan 2017 Reputation: 50 23-Feb-2018, 10:04 PM (This post was last modified: 23-Feb-2018, 10:12 PM by Drone0fPrey.) Yeah RCG get hard to follow sometimes. Might have to stop over if I need some entertaining lol. Its like arguing about, Chevy came out with the spare tire idea first! Ford stole it! A good idea is just that, a good idea. Sooner or later it becomes a standard. In my opinion this hobby would have taken 5-10 more years to get where we are today with sole proprietary development. And the community would be that much smaller too because of it. Sharing is caring! Posts: 331 Threads: 47 Likes Received: 114 in 84 posts Likes Given: 76 Joined: Oct 2017 Reputation: 1 Couldn't find the RCG thread... And I really wanted to laugh too! Lol • |