- linux command
version for many other environments. e.g. cities. Although many of them are for Windows, copy them to related folders.
- remeber to backup a clean version before you start your own version.
export LD_LIBRARY_PATH=/home/jet/anaconda3/lib:$LD_LIBRARY_PATH
work in raw terminal, while not in IPython.
- alter to VS Code & write
~/.bashrc
& run under base -> PASS - install mklml lib -> FAIL
- %env foo=bar, follow the format like other env variables. -> FAIL
- copy
~/anaconda3/lib/libmklml_intel.so
to the other path~/anaconda3/lib/python3.5/site-packages/
, andchmod
-> FAIL - write to
~/.bashrc
-> FAIL - conda base env
- start from (base) prompt -> FAIL
- in ipython
workon
-> FAIL
Both the open source use on Airsim 1.2.0 or earlier, refer to update api for some substitute. Some of them still work with warinings, while some pop erors.
WARNING:root:getPosition API is deprecated. For ground-truth please use simGetGroundTruthKinematics() API.
WARNING:root:getVelocity API is deprecated. For ground-truth please use simGetGroundTruthKinematics() API.
WARNING:root:getCollisionInfo API is renamed to simGetCollisionInfo. Please update your code
SOLUTION
replace any getPosition( )
as simGetGroundTruthKinematics().position
, and getVelocity()
as simGetGroundTruthKinematics().linear_velocity
. as reference from client.py
def getPosition(self):
logging.warning("getPosition API is deprecated. For ground-truth please use simGetGroundTruthKinematics() API." + self.upgrade_api_help)
return self.simGetGroundTruthKinematics().position
def getVelocity(self):
logging.warning("getVelocity API is deprecated. For ground-truth please use simGetGroundTruthKinematics() API." + self.upgrade_api_help)
return self.simGetGroundTruthKinematics().linear_velocity
Also, getCollisionInfo()
is renamed to simGetCollisionInfo()
.
Based on Keras.
- install
theano
,h5py
- run
main.py
of 3D_path_finding/DQN- change path to DQN
- open
main.ipynb
,downloadmain.py
- debug
#from DQN.RL_dataset import Dataset
->from RL_dataset import Dataset
ofRL_dataset.py
Unable to open file (unable to open file: name = 'finalb.h5', errno = 2, error message = 'No such file or directory', flags = 0, o_flags = 0)
-> missfinalb.h5
file. -> disable the read of this file ->msgpackrpc.error.RPCError: rpclib: client error C0002: Function 'getHomeGeoPoint' was called with an invalid number of arguments. Expected: 0, got: 1
- run
main.py
of 3D_path_finding/DDQNImportError: No module named 'cv2'
->conda install -c conda-forge opencv
RPCError: rpclib: client error C0002: Function 'getHomeGeoPoint' was called with an invalid number of arguments. Expected: 0, got: 1
->
- run
test.py
of 3D_path_finding/DDQN captureimages.py
of Object_Detection/Faster_R-CNNRPCError: rpclib: client error C0002: Function 'getHomeGeoPoint' was called with an invalid number of arguments. Expected: 0, got: 1
-> version issue - replaceAirSimClient
asclient
based on CNTK.
-
dqn_tr1.py
(starter)moveToPosition()
->moveToPositionAsync()
goHome()
->goHomeAsync()
-> PASSNameError: name 'AirSimImageType' is not defined
wariningWARNING:root:getVelocity API is deprecated. For ground-truth please use simGetGroundTruthKinematics() API. Please see https://github.com/Microsoft/AirSim/blob/master/docs/upgrade_apis.md for more info. WARNING:root:getPosition API is deprecated. For ground-truth please use simGetGroundTruthKinematics() API. Please see https://github.com/Microsoft/AirSim/blob/master/docs/upgrade_apis.md for more info.
- obstacle setup(TODO)
- not receive api call in UE4 (TODO).
-
NewTestDDQN.py
(official)concated instead of merge ->ValueError: Output tensors to a Model must be the output of a Keras
Layer(thus holding past layer metadata). Found: <keras.layers.merge.Concatenate object at 0x7f6f37298860>
TypeError: 'module' object is not callable
-> py3 grammer of importValueError: Negative dimension size caused by subtracting 8 from 4 for 'conv2d_1/convolution' (op: 'Conv2D') with input shapes: [?,4,84,84], [8,8,84,16].
OR ``- tricks, debug into to see the size,
model.input_size
,model.output_size
after eachConv2D
andMax2Dpooling
. Likely input is unmatched. - refer codes
from keras import backend as K; K.image_data_format()
->'channels_last'
/ input_shape = (img_rows, img_cols, 1) PASS
- tricks, debug into to see the size,
-
ddqn_example.py
(sample code, nothing to do with AirSim)merge()
->TypeError: 'module' object is not callable
- reinstalll tensorflow and keras
concate()
-> size unmatch- can concate substitute the expired merge?
- the CNN, the network match
- print action and reward
- file and data save
Collide with Plane, Car -> set plane coord, Multirotor -> wait
- go on fix -> set the plane coord
copy and work on a new one-> Failstart over comiple-> Fail