feat(exp25): wheel OverlapSphere collision fix + auto-transition
Car.cs (sdsandbox): per-wheel OverlapSphereNonAlloc in FixedUpdate catches barrier contact from any angle, any throttle — forward raycast only covered nose-first. Built, rsync'd, sim restart pending exp24 completion. exp25 script: identical to exp24 params, fresh weights, patched Unity binary. Auto-transition monitor armed: kills sim, restarts with new binary, launches exp25 when exp24 finishes (~22:00 EST). Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
This commit is contained in:
Paul Huliganga
parent
c6a18e7fee
commit
f784fdebd1
|
|
@ -37,6 +37,28 @@ Current runtime simulator folders in use:
|
|||
Unity build log:
|
||||
- `C:\Users\Paul\AppData\Local\Temp\unity_rebuild.log`
|
||||
|
||||
## Car.cs Collision Fix (this session)
|
||||
|
||||
### Per-wheel OverlapSphere (exp25+)
|
||||
|
||||
`/mnt/c/Users/Paul/Documents/projects/sdsandbox/sdsim/Assets/Scripts/Car.cs`
|
||||
|
||||
**Root cause of remaining stuck episodes:** The forward raycast added in the previous
|
||||
session only covered nose-first contact when `requestTorque > 0.05`. Side contact,
|
||||
diagonal contact, rear contact, and low-throttle nose contact were all missed.
|
||||
|
||||
**Fix:** Added per-wheel `OverlapSphere` check in `FixedUpdate()`:
|
||||
- Calls `wc.GetWorldPose()` to get each wheel's world position
|
||||
- Runs `Physics.OverlapSphereNonAlloc(wheelCenter, wc.radius + 0.05f, _wheelOverlapBuf)`
|
||||
- Filters hits by name containing `"barrier"` (matches `left_barrier_seg*`, `right_barrier_seg*`)
|
||||
- Calls `RegisterCollision()` on first barrier hit → episode terminates immediately
|
||||
- Pre-allocated `Collider[8]` buffer as a field (no GC per frame)
|
||||
- Both forward raycast AND OverlapSphere are active (complementary)
|
||||
|
||||
**Build:** Unity 6000.4.4f1, succeeded at ~20:35 on 2026-05-05
|
||||
**Rsync:** Done — runtime folder has new Assembly-CSharp.dll (298,496 bytes)
|
||||
**Effect:** Takes effect after sim restart (pending exp24 completion)
|
||||
|
||||
## What Was Fixed This Session
|
||||
|
||||
### Barrier physics (previous session, still in effect)
|
||||
|
|
@ -115,16 +137,20 @@ to Unity physics non-determinism, NOT road variety.
|
|||
- Rsync'd to: `/mnt/c/Users/Paul/Downloads/DonkeySimWin/DonkeySimWin/`
|
||||
- Sim restarted on port 9091 with new binary
|
||||
|
||||
### Exp 24 status — RUNNING
|
||||
### Exp 24 status — RUNNING (~100 min remaining as of 20:22)
|
||||
- PID 733053, launched at 20:09 on 2026-05-05
|
||||
- Log: `agent/models/exp24-discrete/run_2026-05-05_200903_discrete.log`
|
||||
- Monitor with:
|
||||
```bash
|
||||
tail -f /tmp/exp24.out
|
||||
```
|
||||
NOTE: The log file (run_*_discrete.log) is 0 bytes — logging.basicConfig() is a
|
||||
no-op because gym_donkeycar already configured the root logger. All output goes
|
||||
to /tmp/exp24.out via nohup. Fix in exp25: use `file_log.addHandler(FileHandler(...))` directly.
|
||||
- Monitor: `tail -f /tmp/exp24.out`
|
||||
- Results so far: 10k→250.4r/2000s ✅, 20k→320.9r/2000s ✅ (NEW BEST both times)
|
||||
- NOTE: log file is 0 bytes (logging.basicConfig no-op). All output in /tmp/exp24.out.
|
||||
- Auto-transition to exp25 is armed — when exp24 finishes, monitor kills sim, restarts
|
||||
with wheel-fix binary, and launches exp25 automatically.
|
||||
|
||||
### Exp 25 status — PENDING
|
||||
- Script: `agent/experiments/exp25_wheel_collision_fix.py`
|
||||
- Model dir: `agent/models/exp25-wheel-fix/`
|
||||
- Monitor: `tail -f /tmp/exp25.out` (once launched)
|
||||
- Key fix: per-wheel OverlapSphere in Car.cs catches any-angle barrier contact
|
||||
- Binary: already rsync'd, sim restart needed (auto-transition handles this)
|
||||
|
||||
## Useful Commands
|
||||
|
||||
|
|
|
|||
|
|
@ -0,0 +1,276 @@
|
|||
"""
|
||||
Exp 25: Wheel OverlapSphere collision fix — same setup as exp24, patched Unity sim.
|
||||
|
||||
What changed from exp24:
|
||||
- Unity Car.cs now has per-wheel OverlapSphere checks in FixedUpdate().
|
||||
WheelColliders never fire OnCollisionEnter/Stay on the car body. The previous
|
||||
forward raycast only caught nose-first contact when throttle > 0.05. Now:
|
||||
* Forward raycast: still fires for fast nose-first approach (before contact)
|
||||
* Per-wheel OverlapSphere: fires for any wheel touching any barrier, any angle,
|
||||
regardless of throttle. Filter: only objects with "barrier" in the name.
|
||||
Both checks call RegisterCollision() → episode terminates immediately on contact.
|
||||
- Python speed-check backstop unchanged (speed < 0.5 for 2s → terminate).
|
||||
- Everything else identical to exp24: discrete(7) steering, road regen, LR=0.0003.
|
||||
- Requires sim restart after rsync of new Assembly-CSharp.dll (built 2026-05-05).
|
||||
"""
|
||||
import os
|
||||
import sys
|
||||
import time
|
||||
from datetime import datetime
|
||||
|
||||
sys.path.insert(0, '/home/paulh/projects/donkeycar-rl-autoresearch/agent')
|
||||
|
||||
_SAVE_DIR = '/home/paulh/projects/donkeycar-rl-autoresearch/agent/models/exp25-wheel-fix'
|
||||
_PIDFILE = os.path.join(_SAVE_DIR, 'current.pid')
|
||||
os.makedirs(_SAVE_DIR, exist_ok=True)
|
||||
|
||||
if os.path.exists(_PIDFILE):
|
||||
try:
|
||||
_old = int(open(_PIDFILE).read().strip())
|
||||
if _old != os.getpid():
|
||||
import signal
|
||||
os.kill(_old, 0)
|
||||
print(f'[exp25] Another instance already running (PID {_old}). Exiting.', flush=True)
|
||||
sys.exit(1)
|
||||
except (OSError, ValueError):
|
||||
pass
|
||||
|
||||
import gymnasium as gym
|
||||
import numpy as np
|
||||
from stable_baselines3 import PPO
|
||||
from stable_baselines3.common.vec_env import DummyVecEnv, VecTransposeImage
|
||||
|
||||
from discretize_action import DiscretizedActionWrapper
|
||||
from donkeycar_sb3_runner import ThrottleClampWrapper
|
||||
from multitrack_runner import StuckTerminationWrapper
|
||||
from reward_wrapper import SpeedRewardWrapper
|
||||
|
||||
|
||||
HOST = 'localhost'
|
||||
THROTTLE_MIN = 0.2
|
||||
LR = 0.0003
|
||||
TOTAL_STEPS = 200_000
|
||||
CHECKPOINT_EVERY = 10_000
|
||||
SCENE_RELOAD_WAIT = 5.0
|
||||
|
||||
N_STEER = 7
|
||||
N_THROTTLE = 1
|
||||
|
||||
EFFICIENCY_WINDOW = 30
|
||||
MIN_EFFICIENCY = 0.15
|
||||
MAX_CTE = 8.0
|
||||
MIN_LAP_TIME = 12.0
|
||||
PROGRESS_PATIENCE = 100
|
||||
|
||||
MAX_STUCK_SECONDS = 5.0
|
||||
MAX_EPISODE_SECONDS = 30.0
|
||||
LOW_SPEED_THRESHOLD = 0.5
|
||||
MAX_LOW_SPEED_SECONDS = 2.0
|
||||
|
||||
TRACK_ID = 'donkey-generated-roads-v0'
|
||||
PORT = 9091
|
||||
|
||||
|
||||
def log(msg):
|
||||
print(f'[{datetime.now().strftime("%H:%M:%S")}] {msg}', flush=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 = DiscretizedActionWrapper(env, n_steer=N_STEER, n_throttle=N_THROTTLE)
|
||||
env = StuckTerminationWrapper(
|
||||
env,
|
||||
stuck_steps=40,
|
||||
min_displacement=0.5,
|
||||
max_stuck_seconds=MAX_STUCK_SECONDS,
|
||||
max_episode_seconds=MAX_EPISODE_SECONDS,
|
||||
low_speed_threshold=LOW_SPEED_THRESHOLD,
|
||||
max_low_speed_seconds=MAX_LOW_SPEED_SECONDS,
|
||||
)
|
||||
env = SpeedRewardWrapper(
|
||||
env,
|
||||
window_size=EFFICIENCY_WINDOW,
|
||||
min_efficiency=MIN_EFFICIENCY,
|
||||
max_cte=MAX_CTE,
|
||||
min_lap_time=MIN_LAP_TIME,
|
||||
progress_patience=PROGRESS_PATIENCE,
|
||||
)
|
||||
return env
|
||||
return _init
|
||||
|
||||
|
||||
def connect_env():
|
||||
new_env = DummyVecEnv([make_env(TRACK_ID, PORT)])
|
||||
new_env = VecTransposeImage(new_env)
|
||||
return new_env
|
||||
|
||||
|
||||
def reconnect_env(old_env):
|
||||
try:
|
||||
old_env.close()
|
||||
except Exception as e:
|
||||
log(f' env.close() warning: {e}')
|
||||
time.sleep(SCENE_RELOAD_WAIT)
|
||||
return connect_env()
|
||||
|
||||
|
||||
log('=' * 60)
|
||||
log('Exp 25: generated_road — wheel OverlapSphere collision fix')
|
||||
log(f' Sim: {HOST}:{PORT} -> {TRACK_ID}')
|
||||
log(f' Discrete steering: {N_STEER} bins, throttle fixed at {THROTTLE_MIN}')
|
||||
log(f' LR={LR}, total={TOTAL_STEPS:,}, checkpoint every {CHECKPOINT_EVERY:,}')
|
||||
log(f' Reward: v7 (speed×CTE, efficiency gate, no-progress kill)')
|
||||
log(f' Stuck: position/{MAX_STUCK_SECONDS}s OR speed<{LOW_SPEED_THRESHOLD}/{MAX_LOW_SPEED_SECONDS}s')
|
||||
log(f' Episode cap: {MAX_EPISODE_SECONDS}s | Road regen: every {CHECKPOINT_EVERY:,} steps')
|
||||
log(f' Unity fix: per-wheel OverlapSphere → any-angle barrier detection')
|
||||
log('=' * 60)
|
||||
|
||||
log('Connecting to sim...')
|
||||
env = connect_env()
|
||||
log(f' obs={env.observation_space.shape}, action={env.action_space}')
|
||||
|
||||
model = PPO(
|
||||
'CnnPolicy',
|
||||
env,
|
||||
learning_rate=LR,
|
||||
n_steps=2048,
|
||||
batch_size=64,
|
||||
n_epochs=10,
|
||||
gamma=0.99,
|
||||
gae_lambda=0.95,
|
||||
clip_range=0.2,
|
||||
ent_coef=0.01,
|
||||
verbose=1,
|
||||
device='cpu',
|
||||
)
|
||||
|
||||
with open(_PIDFILE, 'w') as f:
|
||||
f.write(str(os.getpid()))
|
||||
|
||||
log(f'Fresh PPO (Discrete({N_STEER * N_THROTTLE})). Starting training...')
|
||||
|
||||
best_total_steps = float('-inf')
|
||||
best_total_reward = float('-inf')
|
||||
steps_done = 0
|
||||
run_tag = datetime.now().strftime('%Y-%m-%d_%H%M%S') + '_wheel_fix'
|
||||
log_path = os.path.join(_SAVE_DIR, f'run_{run_tag}.log')
|
||||
best_model_path = os.path.join(_SAVE_DIR, 'best_model.zip')
|
||||
|
||||
import logging
|
||||
_file_handler = logging.FileHandler(log_path)
|
||||
_file_handler.setFormatter(logging.Formatter('%(message)s'))
|
||||
_stream_handler = logging.StreamHandler(sys.stdout)
|
||||
_stream_handler.setFormatter(logging.Formatter('%(message)s'))
|
||||
file_log = logging.getLogger('exp25')
|
||||
file_log.setLevel(logging.INFO)
|
||||
file_log.propagate = False
|
||||
file_log.addHandler(_file_handler)
|
||||
file_log.addHandler(_stream_handler)
|
||||
|
||||
|
||||
def flog(msg):
|
||||
ts = datetime.now().strftime('%H:%M:%S')
|
||||
file_log.info(f'[{ts}] {msg}')
|
||||
|
||||
|
||||
flog('=' * 60)
|
||||
flog(f'Exp 25 started — PID {os.getpid()}')
|
||||
flog(f'Log: {log_path}')
|
||||
flog('=' * 60)
|
||||
|
||||
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
|
||||
|
||||
ckpt = os.path.join(_SAVE_DIR, f'checkpoint_{steps_done:07d}')
|
||||
model.save(ckpt)
|
||||
model.save(os.path.join(_SAVE_DIR, 'model'))
|
||||
flog(f'[{steps_done:,}/{TOTAL_STEPS:,}] Checkpoint saved: {ckpt}.zip')
|
||||
|
||||
flog(f' Reconnecting for fresh road...')
|
||||
env = reconnect_env(env)
|
||||
model.set_env(env)
|
||||
flog(f' Connected (new road)')
|
||||
|
||||
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
|
||||
|
||||
total_steps_eval = int(ep_steps.sum())
|
||||
total_reward_eval = float(ep_rewards.sum())
|
||||
|
||||
status = '✅' if ep_steps[0] >= 2000 else f'❌@{int(ep_steps[0])}'
|
||||
flog(f' Eval: gen_road={total_reward_eval:.1f}r/{int(ep_steps[0])}s {status}')
|
||||
|
||||
if (total_steps_eval > best_total_steps
|
||||
or (total_steps_eval == best_total_steps
|
||||
and total_reward_eval > best_total_reward)):
|
||||
best_total_steps = total_steps_eval
|
||||
best_total_reward = total_reward_eval
|
||||
model.save(best_model_path)
|
||||
flog(f' NEW BEST: steps={best_total_steps} reward={best_total_reward:.1f}')
|
||||
|
||||
except Exception as e:
|
||||
flog(f' Eval error: {e}')
|
||||
|
||||
env.close()
|
||||
|
||||
flog('=' * 60)
|
||||
flog('FINAL EVALUATION: best_model on generated_road (3 fresh roads)')
|
||||
flog('=' * 60)
|
||||
|
||||
EVAL_SETS = 3
|
||||
EVAL_MAX_STEPS = 2000
|
||||
|
||||
steps_list = []
|
||||
reward_list = []
|
||||
|
||||
for s in range(1, EVAL_SETS + 1):
|
||||
try:
|
||||
time.sleep(SCENE_RELOAD_WAIT)
|
||||
eval_env = connect_env()
|
||||
eval_model = PPO.load(best_model_path, env=eval_env, device='cpu')
|
||||
obs = eval_env.reset()
|
||||
done = False
|
||||
total_s = 0
|
||||
total_r = 0.0
|
||||
|
||||
while not done and total_s < EVAL_MAX_STEPS:
|
||||
action, _ = eval_model.predict(obs, deterministic=True)
|
||||
result = eval_env.step(action)
|
||||
obs, r, done = result[0], result[1], result[2]
|
||||
if hasattr(done, '__len__'):
|
||||
done = bool(done[0])
|
||||
total_r += float(r) if not hasattr(r, '__len__') else float(r[0])
|
||||
total_s += 1
|
||||
|
||||
status = '✅' if total_s >= EVAL_MAX_STEPS else f'❌@{total_s}'
|
||||
flog(f' Set {s}: {total_r:.1f}r / {total_s}s {status}')
|
||||
steps_list.append(total_s)
|
||||
reward_list.append(total_r)
|
||||
eval_env.close()
|
||||
|
||||
except Exception as e:
|
||||
flog(f' Set {s} error: {e}')
|
||||
|
||||
if steps_list:
|
||||
flog(f' Mean: {np.mean(steps_list):.0f} steps / {np.mean(reward_list):.1f} reward')
|
||||
|
||||
flog('Exp 25 complete.')
|
||||
Loading…
Reference in New Issue