- voting is now at a central place instead of duplicated within the
estimators
-> this also means that estimators that did not do voting so far,
now have voting, like ekf2
- estimators requiring more than that can still subscribe to the raw
sensors
- allows sensors_combined to be 3 times smaller
- reduces logger, memcpy (cache) & RAM overhead
- all modules requiring only 1 or 2 sensor values now automatically get
the voted result
- this also adds voting to baro
Decreases the message size from 780 to 280 bytes.
In particular, all modules using sensor_combined must use the integral now.
The sensor value can easily be reconstructed by dividing with dt.
Voters now need to be moved into sensors module, because error count and
priority is removed from the topic.
Any module that requires additional data from a sensor can subscribe to
the raw sensor topics.
At two places, values are set to zero instead of subscribing to the raw
sensors (with the assumption that no one reads them):
- mavlink mavlink_highres_imu_t::abs_pressure
- sdlog2: sensor temperatures
Drawbacks of the previous method: when writing to the SD card, there are
high delays in the write() call of several 100ms, every now and then. The
frequency and length of these events depend on:
- SD card
- used logger bandwidth
- bandwidth of gps data (RTCM)
Since the whole gps thread was blocked during this period, it lead to
gps timeouts and lost module.
What we do now is: publish an orb topic with queuing. This makes it async
and the logger takes care of buffering. This means it's best to:
- use high logger rate
- use large logger buffer
- reduce logger bandwith by disabling unused topics
Sometimes when flashing new firmware, the IO update fails because safety
is off. In this case, we should set safety on first before putting the
IO board into bootloader mode.
This adds two uORB API calls:
- orb_advertise_queue
- orb_advertise_multi_queue
Both add a queue_size parameter to define a maximum number of buffered
item. The existing orb calls use all a queue size of one and thus their
behavior is unchanged. If a writer publishes too fast, the oldest elements
from the queue are silently dropped.
The returned timestamp is always the one from the latest message in the
queue.
Queue size can be set via ioctl during advertisement phase. After that it
cannot be changed anymore.
Beat Küng