Stand up

An official Microduck move: Pollen Robotics' stand up, retrained from scratch by Microduck Academy with the upstream recipe as-is (task Mjlab-StandUp-Flat-MicroDuck of pollen-robotics/microduck_rl at 2b25a48, 9999 iterations, 4096 envs, no reward edit), so that it carries the base/model.pt checkpoint the Academy needs to remix it. The policy Pollen ships is alpha_stand.onnx.

A utility, not a gallery move. The stand-up alone is a zero-command idle: this repo carries the checkpoint as the base of the Academy's stationary families (a checkpoint to fine-tune from); it wears no gallery tag, so Discover never lists it.

Pollen Robotics' stand-up recipe: from sitting back to the standing pose and holding it (the retrained one recovers from a fall only partly, see the fidelity note).

Fidelity check (this move vs Pollen's policy, same scene, same command, take from_sit)

number this move Pollen's alpha_stand.onnx
final_upright True True
max_tilt 0.139 0.092
base_z_mean 0.115 0.1148
base_z_min 0.0668 0.067
t_to_stand_s 0.2 0.2
z_end 0.1165 0.1163
z_start 0.0696 0.0697
tilt_end 0.005 0.008

Recorded with space/payload/render/rollout.py on the upstream robot MJCF (spikes/base_moves/fidelity.py). PARTIAL fidelity: from SIT the retrained stand-up matches Pollen's (up in 0.2 s, same height, same tilt). From a tipped or post-roll state it settles bowed with the head on the floor (base 6 cm, tilt 0.9) where Pollen's alpha_stand.onnx stands in 0.4 s. Kept as the BASE of the stationary families (a checkpoint to fine-tune from), not as a shipped move; the roulade's chain uses Pollen's stand.

Run it on a robot

sudo robotctl policy add stand-up tfrere/microduck-academy-base-stand
robotctl robot do stand-up

Files

policy.onnx (play it, normalizer baked in), base/model.pt + base/agent.yaml + base/env.yaml (the rsl_rl checkpoint and configs: remix = fine-tune from it), rollouts/0.traj (trajectory.v1 showcase take, with the commands), manifest.json (schema 2 + academy block: origin, source task and policy, training facts, fidelity numbers), fidelity.json, train.json.

Credits and license

The task, the reward design, the robot model and the training recipe are Pollen Robotics' work (pollen-robotics/microduck_rl, Apache-2.0). This repo republishes a retrained policy of that recipe under the same Apache-2.0 license, with the checkpoint, so the Academy's users can start from it.

Downloads last month

-

Downloads are not tracked for this model. How to track
Video Preview
loading