Validation of Human Motor Development Scales with Humanoid Robots

Fanxing Meng; Jing Xiao

cs.ROOct 19, 2026

Validation of Human Motor Development Scales with Humanoid Robots

Authors: Fanxing Meng, Jing Xiao

Organizations: Robotics Engineering Department, Worcester Polytechnic Institute, Worcester, MA, USA

Abstract

We investigate how humanoid robots can learn a series of motor skills defined by human infant motor development scales, and we evaluate which scales are comprehensive enough to serve as self-contained guides for developing gross motor skills up to standing and walking. We represent each motor development scale as a developmental graph that guides the training of a humanoid robot. We then apply this graph within a reinforcement learning framework so that the robot learns skills following the natural phases of human motor development described by the scale. Our work demonstrates that graphs derived from three out of four popular pediatric motor scales can successfully guide humanoid robots to develop these skills. The fourth scale misses some intermediate milestones that the other scales have, which causes some failures. After adding a missing milestone in its graph, it also enables the robot to develop those skills successfully.

Figures & tables

Panel of Figure 1 — Fig. 1: Master graphs derived from each developmental scale. Nodes are topologically sorted; each horizontal row of nodes represents one topological level of the graph. Solid arrow lines indicate that prerequisite milestones need only be achieved once in an episode to activate subsequent nodes.

Panel of Figure 2 — Fig. 2: Mini-graph structures used inside the master graphs. Dashed arrow lines indicate that prerequisite milestones must remain achieved to activate subsequent nodes.

Panel of Figure 3 — Fig. 3: Reproduced examples of visually represented pose milestones in AIMS from [ 1 ] .

Fig. 4: Coordinate systems used in the reward definitions. The world frame is drawn on the lower left, with the vertical $z$ -axis aligned with gravity. The robot’s root reference frame is attached at the pelvis. A link-local frame is shown on the lower leg; its local $\hat{z}$ unit vector (superior direction) projection (purple dashed line) onto the world $z$ -axis is used to compute orientation-based reward term $\hat{z}_{\textrm{lower leg}^{\prime}_{z}}$ .

Description	Expression
Prone
Torso’s anterior axis points downward	$-\hat{x}_{\textrm{torso}_{z}}$
Arm support
Arms’ superior axes point upward	$\hat{z}_{\textrm{arms}_{z}}$
Pivoting
Torso rotates horizontally	$\omega_{\textrm{root}_{z}}$

TABLE I: Definition of standalone node rewards.

Description	Expression
Kneeling with one foot
One foot flat	$\hat{z}_{\textrm{foot}^{\prime}_{z}}+\hat{y}_{\textrm{root}}\cdot\hat{y}_{\textrm{foot}^{\prime}}$
One lower leg upright	$\hat{z}_{\textrm{lower leg}^{\prime}_{z}}$
Other lower leg kneeling	$\hat{y}_{\textrm{lower leg}^{\prime\prime}}\cdot\hat{y}_{\textrm{root}}-\hat{x}_{\textrm{lower leg}^{\prime\prime}_{z}}$
One upper leg horizontal	$\hat{x}_{\textrm{upper leg}^{\prime}_{z}}$
Other upper leg upright	$\hat{z}_{\textrm{upper leg}^{\prime\prime}_{z}}$

TABLE II: Definition of mini-graph reward components.

Simulation		Randomization
Simulation timestep	0.005 s	Observations	$+\mathcal{N}(0,0.002)$
Position iterations	8	Actions	$+\mathcal{N}(0,0.05)$
Velocity iterations	4	Gravity	$+\mathcal{N}(0,0.2)$
Contact offset	0.01 m	Body mass	$\times\mathcal{U}\left[0.8,1.2\right]$
Rest offset	0.0 m	Friction	$\times\mathcal{U}\left[0.7,1.3\right]$
Bounce threshold	0.5 m/s	Damping & stiffness	$\times\mathcal{U}\left[0.8,1.2\right]$

TABLE III: List of hyper-parameters.

Panel of Figure 8 — Fig. 5: Standing-up sequence of robots trained using the four developmental graphs with a 50% passing score. PDMS fails to stand up under this training setting.

Graph	Passing score	Stand success rate	Walking speed (m/s)
AIMS	50%	99%	0.87
	60%	99%	0.96
	70%	98%	0.92
	80%	90%	0.99
BSID	50%	100%	0.95
	60%	99%	1.12

TABLE IV: Stand success rate and walking speed for different graphs and passing scores.

Fig. 6: Total reward for each milestone during training using the GMFM graph with a 50% passing score. Rewards are shown in log scale, suggesting that more advanced milestones dominate learning in later episodes.

	AIMS				BSID				GMFM				PDMS
Passing Score	50%	60%	70%	80%	50%	60%	70%	80%	50%	60%	70%	80%	50%	60%	70%	80%
Prone	0	1	1	1	0	0	1	1	0	0	1	1	1	1	1
Sitting	0	0	0	0	0	0	0	0	0	0	0	1	0	0	0	0
Scooting	–	–	–	–	–	–	–	–	–	–	–	–	0	0	0	1
Arm support	1	2	2	3	1	1	2	6	1	1	2	7	1	1	2
Reaching	1	2	2	4	1	1	2	8	1	1	2	7	1	2

TABLE V: Episode when each milestone is first achieved for different graphs and passing scores.

Fig. 7: Achievement-score heatmap for each milestone and individual component during training using the GMFM graph with a 50% passing score. Alternating dark and light regions indicate that different milestones are optimized at different learning stages.

Milestone	$p_{5}$	$p_{25}$	$p_{50}$	$p_{75}$	$p_{95}$
Sitting	1.1	1.2	1.2	1.3	1.3
Prone	1.5	1.5	1.6	1.7	1.7
Arm support	1.5	1.5	1.6	1.7	1.9
Pivoting	3.8	4.6	5.6	6.4	7.7
Kneeling	1.8	2	2.6	2.8	2.9
Reaching	2.2	2.9

TABLE VI: Time to milestone achievement (seconds).

Robotics

Robotics studies the design and control of machines that perceive, move, and act in physical environments.

See all 755 papers

CHOREO: Every Humanoid Skill as a Trajectory
Sep 11, 2026Ziyi Sun, Jingwen Chen, Yuxi Wang +4Humanoid Robotics Humanoid Robot Control
College of Engineering, Ocean University of China · Faculty of Information Science and Engineering, Ocean University of China · School of Artificial Intelligence, University of Chinese Academy of Sciences +1
Robot Learning from Human Demonstrations: Handwritten Alphabet Trajectories and Human-Likeness Evaluation
Aug 6, 2026Alperen Kenan, Paul Bremner, Manuel GiulianiRobot Skill Learning Gaussian Mixture Models
Bristol Robotics Laboratory, University of the West of England, Bristol, United Kingdom · Faculty of Electrical Engineering, Kempten University of Applied Sciences, Kempten, Germany
Developing Combined Manipulation and Locomotion Skills with Interaction Representation and Skill Composition
Jul 31, 2026Fanxing Meng, Jing XiaoRobotic Grasping Robot Skill Learning
Robotics Engineering Department, Worcester Polytechnic Institute, Worcester, MA, USA

Sep 11, 2026J/K move · Enter open · S save

CHOREO: Every Humanoid Skill as a Trajectory

Ziyi Sun, Jingwen Chen, Yuxi Wang, Xiuze Xia+3

College of Engineering, Ocean University of China · Faculty of Information Science and Engineering, Ocean University of China · School of Artificial Intelligence, University of Chinese Academy of Sciences+1

Humanoid Robotics Humanoid Robot Control Robot Motion Generation Robot Skill Learning

Open paper AskPDF arXiv

Recent advances in humanoid robotics have produced diverse skills through reinforcement learning, motion imitation, and generative modeling. Yet these capabilities remain siloed because they are built around incompatible representations, interfaces, and controllers. We present CHOREO, a framework for training-free composition of heterogeneous humanoid skills. Our key observation is that, regardless of how a skill is learned, it can ultimately be expressed as an executable motion trajectory. Based on this observation, CHOREO converts each capability into SkillMotion, a unified representation that combines motion states, contacts, semantics, and boundary conditions. Skills are composed through direct continuation, cubic Hermite blending, or validated bridge motions, without retraining source models or updating models at test time. On Unitree G1 in MuJoCo, CHOREO organizes 2,950 admitted SkillMotion assets derived from heterogeneous sources and achieves 95.4% sequence success across 130 multi-action tasks, including 93.8% success on eight-action sequences. These results demonstrate that executable trajectories provide a scalable interface for accumulating and composing pretrained humanoid capabilities.

Abstract

Robotics

Explore similar work

CHOREO: Every Humanoid Skill as a Trajectory

Robot Learning from Human Demonstrations: Handwritten Alphabet Trajectories and Human-Likeness Evaluation

Developing Combined Manipulation and Locomotion Skills with Interaction Representation and Skill Composition