omkarpatil/ffw_sg2_wave-right_diffusion_state

FFW SG2 · wave-right · proprioception-only Diffusion Policy

Diffusion Policy (LeRobot 0.6.1) trained on the wave-right task of omkarpatil/wave-traj (15 teleop episodes, ROBOTIS AI Worker ffw_sg2_rev1, instruction "wave using the right hand"). The policy conditions on joint state only — no cameras.

Inputs	observation.state (22): arm_l ×7, gripper_l, arm_r ×7, gripper_r, head ×2, lift, cmd_vel linear_x / linear_y / angular_z
Outputs	action (22), same layout, published to /leader/*/joint_trajectory and /cmd_vel
Chunking	n_obs_steps=1, horizon=32, n_action_steps=16 at 15 Hz (= 2.1 s predicted, 1.07 s executed per chunk)
Inference	DDPM, num_inference_steps=10 (≈ 70 ms on an A5000; 100 steps gives the same accuracy at ~650 ms)
Training	30 k steps, batch 64, lr 1e-4 cosine, seed 1000; final loss 1e-3; offline 16-step open-loop MAE 0.002 rad

Initial state for inference

The demonstrations all start from the pose below (mean over 15 episodes; std is the spread across demos). Put the robot at this pose before issuing START — the policy has never seen states far from it, and in MuJoCo rollouts it waved reliably from starts up to ~0.24 rad (per joint) away from the nearest demo start. Head and lift were fixed during collection, so use exactly those values. Base velocity dims must read ~0.

joint	mean [rad]	std	range over demos
arm_l_joint1	-0.138	0.061	[-0.222, -0.036]
arm_l_joint2	+0.168	0.027	[+0.114, +0.203]
arm_l_joint3	-0.081	0.052	[-0.180, -0.004]
arm_l_joint4	-1.452	0.112	[-1.669, -1.298]
arm_l_joint5	+0.198	0.054	[+0.104, +0.281]
arm_l_joint6	+0.079	0.096	[-0.126, +0.233]
arm_l_joint7	+0.018	0.034	[-0.037, +0.109]
gripper_l_joint1	+0.254	0.063	[+0.163, +0.345]
arm_r_joint1	-0.161	0.079	[-0.304, -0.045]
arm_r_joint2	-0.069	0.024	[-0.106, -0.024]
arm_r_joint3	+0.025	0.056	[-0.075, +0.142]
arm_r_joint4	-1.402	0.125	[-1.627, -1.232]
arm_r_joint5	-0.026	0.047	[-0.115, +0.053]
arm_r_joint6	-0.012	0.093	[-0.206, +0.113]
arm_r_joint7	-0.015	0.119	[-0.252, +0.130]
gripper_r_joint1	+0.258	0.161	[+0.054, +0.503]
head_joint1	-0.204	0.000	[-0.204, -0.204]
head_joint2	-0.292	0.000	[-0.292, -0.292]
lift_joint	-0.000	0.001	[-0.002, +0.000]
linear_x	-0.000	0.000	[-0.000, +0.000]
linear_y	+0.000	0.000	[-0.000, +0.000]
angular_z	+0.000	0.000	[-0.001, +0.000]

The same numbers are in initial_state.json (initial_state_mean is the vector to command, in joint_names order; final_state_mean is where the demos end, i.e. the rest pose the policy returns to).

Running it with cyclo_intelligence

• Set the task's inference_hz = 15 (default) and control_hz = 100 in the UI; ActionChunkProcessor spaces the chunk's 16 steps at 1/inference_hz, so 15 Hz must match the dataset fps this policy was trained at.
• Stock LeRobot refuses to build a Diffusion Policy without an image/environment-state input, and the container engine sends a single (B, D) state per request. Both are handled by cyclo_brain/policy/lerobot/lerobot_engine/diffusion_compat.py (loaded automatically by the engine's loading.py / prediction.py) — the policy container needs that version of the bind-mounted lerobot_engine/. Loading in plain LeRobot: call diffusion_compat.allow_state_only_diffusion() first, then DiffusionPolicy.from_pretrained(...), and add the time axis with expand_obs_time_dim(batch, 1) before predict_action_chunk.

MuJoCo rollouts

10/10 full waves from sampled initial states in the lerobot-mujoco-tutorial FFW SG2 model (right-arm joint range 88 % of the demonstrations', left arm quiet, 0.012 rad servo tracking error).