diff --git a/agent/experiments/exp11_parallel_envs.py b/agent/experiments/exp11_parallel_envs.py
new file mode 100644
index 0000000..077413b
--- /dev/null
+++ b/agent/experiments/exp11_parallel_envs.py
@@ -0,0 +1,129 @@
+"""
+Exp 11: Parallel DummyVecEnv — generated_track + mountain_track on two sim instances.
+
+THIS IS THE KEY EXPERIMENT: Does parallel multi-track training via DummyVecEnv
+produce reliable results, unlike the close-and-switch approach (Wave 4, Exp 10)?
+
+Setup:
+  - Sim 1: 10.0.0.55:9091 → generated_track
+  - Sim 2: 10.0.0.55:9093 → mountain_track
+  - DummyVecEnv wraps both → PPO sees both tracks in every rollout batch
+  - NO env closing, NO set_env(), NO track switching
+  - Same hyperparameters as Wave 4 Trial 9 (the lottery winner)
+
+Hypothesis: Parallel envs eliminate catastrophic forgetting. Results should be
+CONSISTENT across runs, not a 16% lottery like Wave 4.
+"""
+import sys, os, time
+sys.path.insert(0, '/home/paulh/projects/donkeycar-rl-autoresearch/agent')
+
+from multitrack_runner import log, StuckTerminationWrapper
+from donkeycar_sb3_runner import ThrottleClampWrapper
+from reward_wrapper import SpeedRewardWrapper
+from stable_baselines3 import PPO
+from stable_baselines3.common.vec_env import DummyVecEnv, VecTransposeImage
+import gymnasium as gym
+import numpy as np
+
+HOST         = '10.0.0.55'
+THROTTLE_MIN = 0.2
+LR           = 0.000725
+TOTAL_STEPS  = 90000
+SAVE_DIR     = '/home/paulh/projects/donkeycar-rl-autoresearch/agent/models/exp11-parallel-envs'
+os.makedirs(SAVE_DIR, exist_ok=True)
+
+def make_env(track_id, port):
+    def _init():
+        raw = gym.make(track_id, conf={'host': HOST, 'port': port})
+        env = ThrottleClampWrapper(raw, throttle_min=THROTTLE_MIN)
+        env = StuckTerminationWrapper(env, stuck_steps=80, min_displacement=0.5)
+        env = SpeedRewardWrapper(env)
+        return env
+    return _init
+
+log('='*60)
+log('Exp 11: Parallel DummyVecEnv — two sims, two tracks')
+log(f'  Sim 1: {HOST}:9091 → generated_track')
+log(f'  Sim 2: {HOST}:9093 → mountain_track')
+log(f'  throttle_min={THROTTLE_MIN}, lr={LR}, total={TOTAL_STEPS:,}')
+log(f'  Method: DummyVecEnv (both tracks in every PPO batch)')
+log(f'  NO close_and_switch — single stable env for entire training')
+log('='*60)
+
+# Create parallel env
+log('Creating DummyVecEnv with two tracks...')
+env = DummyVecEnv([
+    make_env('donkey-generated-track-v0', 9091),
+    make_env('donkey-mountain-track-v0', 9093),
+])
+env = VecTransposeImage(env)
+log(f'  VecEnv num_envs={env.num_envs}, obs={env.observation_space.shape}')
+
+# Create PPO
+model = PPO('CnnPolicy', env, learning_rate=LR, verbose=1, device='cpu')
+log('PPO created. Starting training...')
+
+# Train in segments for checkpointing
+CHECKPOINT_EVERY = 6000
+best_reward = float('-inf')
+steps_done = 0
+
+while steps_done < TOTAL_STEPS:
+    seg_steps = min(CHECKPOINT_EVERY, TOTAL_STEPS - steps_done)
+    model.learn(total_timesteps=seg_steps, reset_num_timesteps=False)
+    steps_done += seg_steps
+
+    # Save checkpoint
+    ckpt = os.path.join(SAVE_DIR, f'checkpoint_{steps_done:07d}')
+    model.save(ckpt)
+    log(f'[{steps_done:,}/{TOTAL_STEPS:,}] Checkpoint saved: {ckpt}.zip')
+
+    # Quick eval on the parallel env (both tracks)
+    try:
+        obs = env.reset()
+        ep_rewards = np.zeros(env.num_envs)
+        ep_steps = np.zeros(env.num_envs)
+        done_mask = np.zeros(env.num_envs, dtype=bool)
+        for _ in range(2000):
+            action, _ = model.predict(obs, deterministic=True)
+            obs, rewards, dones, infos = env.step(action)
+            for i in range(env.num_envs):
+                if not done_mask[i]:
+                    ep_rewards[i] += rewards[i]
+                    ep_steps[i] += 1
+                    if dones[i]:
+                        done_mask[i] = True
+            if done_mask.all():
+                break
+        log(f'  Eval: env0(gen_track)={ep_rewards[0]:.1f}r/{int(ep_steps[0])}steps  '
+            f'env1(mountain)={ep_rewards[1]:.1f}r/{int(ep_steps[1])}steps')
+
+        total_reward = ep_rewards.sum()
+        if total_reward > best_reward:
+            best_reward = total_reward
+            model.save(os.path.join(SAVE_DIR, 'best_model'))
+            log(f'  ⭐ NEW BEST: {best_reward:.1f} (combined)')
+    except Exception as e:
+        log(f'  Eval error: {e}')
+
+model.save(os.path.join(SAVE_DIR, 'model'))
+env.close()
+time.sleep(3)
+
+log(f'\nTraining complete. Best combined reward: {best_reward:.1f}')
+log(f'Checkpoints in {SAVE_DIR}:')
+for f in sorted(os.listdir(SAVE_DIR)):
+    size = os.path.getsize(os.path.join(SAVE_DIR, f)) // (1024*1024)
+    log(f'  {f}  ({size}MB)')
+
+# Run standard eval
+log('\nRunning standard 3-set eval on all tracks...')
+import subprocess
+subprocess.run([
+    'python3',
+    '/home/paulh/projects/donkeycar-rl-autoresearch/agent/run_eval.py',
+    '--model', os.path.join(SAVE_DIR, 'best_model.zip'),
+    '--sets', '3', '--steps', '2000'
+], cwd='/home/paulh/projects/donkeycar-rl-autoresearch/agent')
+
+log('\n=== Exp 11 COMPLETE ===')