File size: 7,759 Bytes
3f4bb1d | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 | #!/usr/bin/env python3
"""SONIC walks to the WalkGrab object on a planner reference, then reaches.
The action term tracks the GEAR planner clip in SONIC's ``g1`` mode and hands
over to wrist teleop by itself, so this script just holds a neutral action
while it walks and then runs the same reach / close / lift schedule as the
standing manip smoke.
Generate the reference first (matched to the object xy):
./jobs/gen_walk_reference.sh --target 1.5 -0.35
./jobs/sonic_walkgrab_play.sh --headless --video
"""
from __future__ import annotations
import argparse
import os
import pathlib
import sys
import traceback
_PROJECT = pathlib.Path(__file__).resolve().parents[2]
_SIM_SRC = _PROJECT / "simulation" / "source"
if str(_SIM_SRC) not in sys.path:
sys.path.insert(0, str(_SIM_SRC))
from isaaclab.app import AppLauncher
parser = argparse.ArgumentParser(
description="SONIC whole-body walk to the WalkGrab object, then wrist reach."
)
parser.add_argument("--task", default="V2D-G1-SonicWalkGrab-Play-v0")
parser.add_argument(
"--clip",
default="20200709_141754_836212060125",
help="reconstruction/runs/<clip> whose exported object.urdf goes on the table.",
)
parser.add_argument("--num_envs", type=int, default=1)
parser.add_argument("--max_steps", type=int, default=0, help="0 = one full episode.")
parser.add_argument("--video", action="store_true")
parser.add_argument(
"--video_length",
type=int,
default=0,
help="Recorded steps; 0 = whole run. Used when --video is set.",
)
parser.add_argument(
"--reach-xy",
type=float,
default=None,
help="Override the pelvis-object xy range at which the term stops tracking the "
"planner reference and hands over to the wrist.",
)
AppLauncher.add_app_launcher_args(parser)
args_cli = parser.parse_args()
os.environ["V2D_SONIC_WALKGRAB_CLIP"] = args_cli.clip
if args_cli.video:
args_cli.enable_cameras = True
app_launcher = AppLauncher(args_cli)
simulation_app = app_launcher.app
import gymnasium as gym # noqa: E402
import torch # noqa: E402
from isaaclab.utils.dict import print_dict # noqa: E402
from isaaclab_tasks.utils import parse_env_cfg # noqa: E402
import v2d_sim.tasks.g1_sonic_walkgrab # noqa: E402, F401
_GRAV, _WRIST, _OBJ, _DELTA = (
slice(0, 3),
slice(3, 6),
slice(6, 9),
slice(9, 12),
)
def _t(x):
return x.torch if hasattr(x, "torch") else x
def main() -> None:
env_cfg = parse_env_cfg(args_cli.task, device=args_cli.device, num_envs=args_cli.num_envs)
if args_cli.reach_xy is not None:
env_cfg.actions.wrist.arrive_xy = args_cli.reach_xy
env = gym.make(
args_cli.task,
cfg=env_cfg,
render_mode="rgb_array" if args_cli.video else None,
)
episode_steps = int(env.unwrapped.max_episode_length)
n = args_cli.max_steps or episode_steps
video_length = args_cli.video_length or n
print(
f"[sonic-walkgrab] task={args_cli.task} clip={args_cli.clip} "
f"episode={episode_steps} running={n} recording={video_length}",
flush=True,
)
print(
"[sonic-walkgrab] SONIC g1 mode tracks the GEAR planner walk reference, then "
f"hands over to wrist teleop inside {env_cfg.actions.wrist.arrive_xy:.2f} m pelvis xy.",
flush=True,
)
if args_cli.video:
out_dir = _PROJECT / "simulation" / "runs" / "sonic_walkgrab" / args_cli.clip
out_dir.mkdir(parents=True, exist_ok=True)
video_kwargs = {
"video_folder": str(out_dir),
"step_trigger": lambda step: step == 0,
"video_length": video_length,
"disable_logger": True,
}
print("[sonic-walkgrab] recording video", flush=True)
print_dict(video_kwargs, nesting=4)
env = gym.wrappers.RecordVideo(env, **video_kwargs)
inner = env.unwrapped
wrist_term = inner.action_manager.get_term("wrist")
acfg = env_cfg.actions.wrist
lo = torch.tensor(
[acfg.wrist_x[0], acfg.wrist_y[0], acfg.wrist_z[0]], device=inner.device
)
hi = torch.tensor(
[acfg.wrist_x[1], acfg.wrist_y[1], acfg.wrist_z[1]], device=inner.device
)
center, half = (lo + hi) / 2.0, (hi - lo) / 2.0
obs_dict, _ = env.reset()
obs = obs_dict["policy"]
robot = inner.scene["robot"]
obj = inner.scene["object"]
rest_z = float(env_cfg.scene.object.init_state.pos[2])
n_env = inner.num_envs
reach_start = torch.full((n_env,), -1, device=inner.device, dtype=torch.long)
min_root_z = float("inf")
min_hand_obj = float("inf")
max_root_x = float("-inf")
max_lift = 0.0
settle = 10
obj_z0 = rest_z
hold = torch.zeros(n_env, 4, device=inner.device)
hold[:, 2] = 1.0
hold[:, 3] = -1.0
for step in range(n):
if not simulation_app.is_running():
break
obj_xy = torch.linalg.norm(obs[:, _OBJ][:, :2], dim=1)
# The action term owns the handover; mirror it so the reach schedule
# starts on the same step SONIC switches out of g1 mode.
arrived = torch.as_tensor(~wrist_term.walking, device=inner.device)
reach_start[(reach_start < 0) & arrived] = step
in_reach = reach_start >= 0
action = hold.clone()
if bool(in_reach.any()):
local = (step - reach_start.clamp(min=0)).float()
horizon = max(1.0, float(n - int(reach_start[in_reach].min())))
frac = (local / horizon).clamp(0.0, 1.0)
want = obs[:, _OBJ].clone()
want[:, 2] = want[:, 2] + 0.06 * (1.0 - frac)
want[:, 2] = want[:, 2] + torch.where(
frac > 0.60, 0.12 * (frac - 0.60) / 0.40, torch.zeros_like(frac)
)
grip = torch.where(
frac < 0.45,
torch.full_like(frac, -1.0),
torch.where(
frac < 0.60,
-1.0 + 2.0 * (frac - 0.45) / 0.15,
torch.ones_like(frac),
),
)
a = torch.clamp((want - center) / half, -1.0, 1.0)
reach_action = torch.cat([a, grip.unsqueeze(-1)], dim=1)
action[in_reach] = reach_action[in_reach]
obs_dict, _, terminated, truncated, _ = env.step(action)
obs = obs_dict["policy"]
root = _t(robot.data.root_pos_w)[0]
obj_pos = _t(obj.data.root_pos_w)[0]
min_root_z = min(min_root_z, float(root[2]))
max_root_x = max(max_root_x, float(root[0]))
if step == settle:
obj_z0 = float(obj_pos[2])
max_lift = max(max_lift, float(obj_pos[2]) - obj_z0)
hand_obj = float(torch.linalg.norm(obs[0, _DELTA]))
min_hand_obj = min(min_hand_obj, hand_obj)
if step % 50 == 0 or step == n - 1:
mode = "reach" if bool(in_reach[0]) else "walk"
print(
f" step {step}/{n} {mode} root_xy=({float(root[0]):.3f},{float(root[1]):.3f}) "
f"root_z={float(root[2]):.3f} object_xy=({float(obj_pos[0]):.3f},{float(obj_pos[1]):.3f}) "
f"pelvis-obj_xy={float(obj_xy[0]):.3f} hand-obj={hand_obj:.3f} "
f"done={int(bool(terminated[0] or truncated[0]))}",
flush=True,
)
print(f"[sonic-walkgrab] min root_z={min_root_z:.3f} max root_x={max_root_x:.3f}", flush=True)
print(f"[sonic-walkgrab] min hand-object (pelvis frame)={min_hand_obj:.3f} m", flush=True)
print(f"[sonic-walkgrab] peak object lift above rest={max_lift:+.3f} m", flush=True)
env.close()
if __name__ == "__main__":
code = 0
try:
main()
except Exception:
traceback.print_exc()
code = 1
finally:
simulation_app.close()
os._exit(code)
|