/terminal-guidance-law

design of terminal guidance law based on sparse reward reinforce learning

Primary LanguagePython

DESIGN OF TERMINAL GUIDANCE LAW BASED ON SPARSE REWARD REINFORCE LEARNING

Prioritized Experience Replay (PER)

airdensity.py : about aerodynamics, from DOI:10.27061/d.cnki.ghgdu.2021.003352

table.py : about aerodynamics, from DOI:10.27061/d.cnki.ghgdu.2021.003352

testEnv.py : environment, from DOI:10.27061/d.cnki.ghgdu.2021.003352

dqn_train.py : run this to train DQN terminal guidance law, from DOI:10.27061/d.cnki.ghgdu.2021.003352

dqn_test.py : run this to test DQN terminal guidance law, from DOI:10.27061/d.cnki.ghgdu.2021.003352

per_sumtree.py : sumtree, from Howuhh/prioritized_experience_replay

per_buffer.py : per buffer, from Howuhh/prioritized_experience_replay

per_train.py : run this to train PER terminal guidance law, references Howuhh/prioritized_experience_replay

per_test.py : run this to test PER terminal guidance law,

Results

result

References

environment from DOI : 10.27061/d.cnki.ghgdu.2021.003352

PER from : Howuhh/prioritized_experience_replay

Hindsight Experience Replay (HER)

airdensity.py : about aerodynamics, from DOI:10.27061/d.cnki.ghgdu.2021.003352

table.py : about aerodynamics, from DOI:10.27061/d.cnki.ghgdu.2021.003352

testEnv.py : environment, from DOI:10.27061/d.cnki.ghgdu.2021.003352

hindsightEnv.py : to “hindsight” calculate state using the trajectory of the missile and the target

hindsight_dqn_train.py : run this to train Hindsight terminal guidance law

hindsight_dqn_test.py : run this to test Hindsight terminal guidance law

hindsight_per_train.py : can only get bad results, perhaps the priorities of hindsight transition should not be calculated using TD error