Fix the buffer bug and add td3 for elegantrl

rofunc/examples/learning/CURICabinet_elegantrl.py

@@ -28,15 +28,25 @@ def eval(custom_args, ckpt_path=None):
     # TODO: add support for eval mode
     beauty_print("Start evaluating")
 
-    env, args = setup(custom_args, eval_mode=True)
+    env, agent = setup(custom_args, eval_mode=True)
 
     # load checkpoint
     if ckpt_path is None:
         ckpt_path = model_zoo(name="CURICabinetPPO_right_arm.pt")
-    agent.load(ckpt_path)
+    agent.save_or_load_agent(cwd=ckpt_path, if_save=False)
 
     # evaluate the agent
-    trainer.eval()
+    state = env.reset()
+    episode_reward = 0
+    for i in range(2 ** 10):
+        action = agent.act.get_action(state).detach()
+        next_state, reward, done, _ = env.step(action)
+        episode_reward += reward.mean()
+        # if done:
+        #     print(f'Step {i:>6}, Episode return {episode_reward:8.3f}')
+        #     break
+        # else:
+        state = next_state
 
 
 if __name__ == '__main__':
@@ -47,15 +57,12 @@ def eval(custom_args, ckpt_path=None):
     parser.add_argument("--sim_device", type=str, default="cuda:{}".format(gpu_id))
     parser.add_argument("--rl_device", type=str, default="cuda:{}".format(gpu_id))
     parser.add_argument("--graphics_device_id", type=int, default=gpu_id)
-    parser.add_argument("--headless", type=str, default="False")
+    parser.add_argument("--headless", type=str, default="True")
     parser.add_argument("--test", action="store_true", help="turn to test mode while adding this argument")
     custom_args = parser.parse_args()
 
     if not custom_args.test:
         train(custom_args)
     else:
-        # TODO: add support for eval mode
-        folder = 'CURICabinetSAC_22-11-27_18-38-53-296354'
-        ckpt_path = "/home/ubuntu/Github/Knowledge-Universe/Robotics/Roadmap-for-robot-science/rofunc/examples/learning/runs/{}/checkpoints/best_agent.pt".format(
-            folder)
+        ckpt_path = "/home/ubuntu/Github/Knowledge-Universe/Robotics/Roadmap-for-robot-science/rofunc/examples/learning/result/CURICabinet_SAC_42/actor_53608448_00007.742.pth"
         eval(custom_args, ckpt_path=ckpt_path)

rofunc/lfd/rl/utils/elegantrl_utils.py

@@ -12,9 +12,12 @@
 from elegantrl.train.config import Arguments
 from elegantrl.agents.AgentPPO import AgentPPO
 from elegantrl.agents.AgentSAC import AgentSAC
+from elegantrl.agents.AgentTD3 import AgentTD3
+from elegantrl.train.run import init_agent
 
 from rofunc.config.utils import get_config, omegaconf_to_dict
 from rofunc.lfd.rl.tasks import task_map
+from rofunc.utils.logger.beauty_logger import beauty_print
 
 
 class ElegantRLIsaacGymEnvWrapper:
@@ -96,6 +99,7 @@ def step(
 def setup(custom_args, eval_mode=False):
     # get config
     sys.argv.append("task={}".format(custom_args.task))
+    beauty_print("Agent: {}{}ElegantRL".format(custom_args.task, custom_args.agent.upper()), 2)
     sys.argv.append("sim_device={}".format(custom_args.sim_device))
     sys.argv.append("rl_device={}".format(custom_args.rl_device))
     sys.argv.append("graphics_device_id={}".format(custom_args.graphics_device_id))
@@ -106,6 +110,7 @@ def setup(custom_args, eval_mode=False):
 
     if eval_mode:
         task_cfg_dict['env']['numEnvs'] = 16
+        cfg.headless = False
 
     env = task_map[custom_args.task](cfg=task_cfg_dict,
                                      rl_device=cfg.rl_device,
@@ -121,6 +126,8 @@ def setup(custom_args, eval_mode=False):
         agent_class = AgentPPO
     elif custom_args.agent.lower() == "sac":
         agent_class = AgentSAC
+    elif custom_args.agent.lower() == "td3":
+        agent_class = AgentTD3
     else:
         raise ValueError("Agent not supported")
 
@@ -143,4 +150,8 @@ def setup(custom_args, eval_mode=False):
     args.learner_gpus = cfg.graphics_device_id
     args.random_seed = 42
 
+    if eval_mode:
+        agent = init_agent(args, args.learner_gpus, env)
+        return env, agent
+
     return env, args