Most robotic handwriting rigidly mounts the pen and reduces writing to trajectory tracking. A freely held pen instead couples grasp stability, tip motion, and contact force. We present Grip, Trace, Write, a force-aware simulation environment and learning framework for high-DoF handwriting. A two-branch actor separates pen retention from arm-wrist writing while preserving their closed-loop coupling. Training learns a grasp controller, initializes writing from inverse-kinematics demonstrations by behavior cloning, and improves writing through reinforcement learning with rewards on ink, force, time, and retention. Across diverse tasks, the policy improves completion, precision, force regulation, and efficiency over behavior cloning and inverse kinematics, and generalizes without fine-tuning to unseen English and Chinese content. These results establish force-aware handwriting as a testbed for dexterous tool use.
Figures & tables
Figure 1: Hand kinematics and two-branch actor. (a)(b) Spheres mark the finger joints, colored by finger, with rods along the hinge axes; the skin is rendered translucent. The hand has 25 finger joints, in addition to 2 wrist and 7 arm joints. (c) Actor model reads both the same state and trajectory. A fixed output mask selects finger actions from the grasp branch and UR5e arm, wrist, and stop actions from the writing branch. The branches interact through the held pen and live state.
Figure 2: Training pipeline. Panels 1–3 correspond to the three stages in Sec. 2.3 : grasp BC and hold RL, all-task writing BC with the learned grasp, and full-task outcome RL on the composed policy.
Controller
Metric
Primary tasks
Unseen
Mean
Line
Circle
Shape
Digits
Letters
Word
Cursive
Chinese
word
Idiom
Chinese
IK
Completion (%) ↑
98.2
98.8
94.9
90.6
92.7
85.4
83.4
91.7
87.7
92.1
87.7
92.0
Precision (%) ↑
95.7
96.4
82.6
75.8
82.2
71.5
61.6
78.7
71.1
82.1
71.7
80.6
Force MAE (N) ↓
0.558
0.537
0.529
0.531
0.533
0.582
0.526
0.497
0.585
0.580
0.524
0.537
Time (s) ↓
3.16
4.47
7.38
5.05
5.91
22.27
8.48
8.19
18.96
27.71
12.44
8.11
BC
Completion (%) ↑
96.7
97.6
93.6
90.2
92.5
85.7
82.4
91.5
87.2
91.6
86.9
91.3
Table 1: Task-level and zero-shot results with 50 matched seeds per task. Completion, precision, force MAE, and time are reported on separate rows. The mean covers the eight primary tasks; the three unseen columns are zero-shot evaluations. Bold marks the best controller for each task and metric.
Appendix figures & tables13 assets
Supplementary material from the paper’s appendix.
Appendix
Parameter
Hand and environment configuration
Hand and arm
High-DoF anthropomorphic robotic hand mounted on a UR5e assembly
Actuated coordinates
Seven arm coordinates, two wrist joints, and 25 finger joints
Free rigid-body pen, initially seated in a tripod grasp; retained through contact
Grip assistance
No virtual grip assistance in the reported writing configuration
Appendix
Table 2: Hand, environment, and task-interface settings for the reported writing study. These specify the physical control problem; model architecture and optimization settings are given separately in Table 3 .
Figure 3: Simulated setup. (a) A UR5e arm with a 2-DoF wrist and a 25-DoF anthropomorphic hand writes on paper with a free pen. (b) The tripod grasp. (c) Top view mid-episode; gray marks the remaining target.
Parameter
Model, training, and evaluation configuration
Model architecture and observation processing
Input representation
137 state entries plus 16 trajectory samples of five entries each; samples every two trajectory indices; no image encoder
Actor
Separate grasp and writing MLPs, each 217→256→256 with tanh activations; masked linear heads
Policy output
34 physical-action entries plus one learned stop output; stop threshold 0.5
Critic
Independent 217→256→256→1 MLP
Normalization
Vector statistics frozen during final RL; normalized observations clipped to [−10,10]
Appendix
Table 3: Model and learning settings, separated from the hand/environment parameters in Table 2 . BC settings describe the available collection/training recipe, not a reconstructed per-shard manifest of the final clone. Evaluation settings specify measurement rather than training updates.
Component
Training condition
Training and transfer
Grasp controller
240 teacher episodes for BC, then pen-hold PPO under motion and contact
Transfer the selected learned grasp into the MLP
Writing data and BC
Traces, glyphs, words/cursive, and Chinese together
IK arm/wrist plus learned grasp supplies clean/noisy demonstrations; fit one all-task clone
Composed-policy RL
Full writing mixture with grasp and contact dynamics active
Both branches execute; refine outcomes while retaining grasp-mean weights
Appendix
Table 4: Transferred grip controller, all-task writing demonstration collection and BC, and full-task composed-policy RL.
Figure 4: Training across the pipeline. (a) Pen-hold RL from the BC grasp: fraction of training episodes that keep the pen (thin: per iteration; thick: 15-iteration mean) and 24-seed evaluations over static hold, moving hold, and scribble. (b) Per-task completion of all-task BC and its IK teacher (50 seeds). (c, d) Writing RL completion and force MAE over environment steps; BC at step 0, IK dashed.
Weight
Value
Reward term
λC
1000
Ordered target coverage
λP
500
Off-target ink
λE
200
Contact-force MAE
λτ
300
Normalized elapsed time
λq
5000
Progress toward the guidance point
Appendix
Table 5: Weights in the final drawing-outcome reward.
Task ID
Nominal duration
Requested behavior and configuration
L0: grasp training; arm/wrist motion supplied by the environment
L0_hold_static
6 s
Retain the seated pen while the hand is still; no target ink strokes.
L0_hold_moving
8 s
Retain the pen during imposed transport and wrist disturbance: 11 waypoints, motion scale 2.2, arm-motion box half-extents (0.12,0.08,0.085) m, wrist shake amplitude 4∘ .
L0_scribble
10 s
Retain the pen during an environment-driven paper-contact path: 14 waypoints, 3 mm commanded press depth, 8∘ wrist wobble, and 15 N impact limit. This is a contact-retention task, not glyph matching.
L1: geometric tracing
L1_trace_line
8 s
A 60–120 mm straight line with randomized orientation and placement; flat force profile.
Appendix
Table 6: Task catalogue relevant to this paper, from the current samplers. Dimensions are requested geometric ranges before width fitting. Nominal duration paces the trajectory; it is not necessarily the final elapsed episode time. All listed L1/L2/E2 writing samplers use the pen configuration.
Figure 5: Task suite. Top: the three pen-retention tasks. Bottom: one target per tracing and writing task, drawn to scale with its size; * marks held-out content evaluated zero-shot.
Field
Type / unit
Meaning
task_id , level
Identifier
Select the task family and level.
instruction
Text
Human-readable description. The policy MLP does not encode this string as language.
strokes
List of Nj×2 arrays, m
Ordered centerline points in paper coordinates relative to the sheet center. List boundaries distinguish pen-down strokes.
forces
List of Nj arrays, N
Requested pressing force at each corresponding stroke point.
widths
List of Nj arrays, mm
Target mark widths derived from the force arrays and the tool’s ink model.
duration
Seconds
Nominal schedule duration used to pace the requested strokes and transits.
Appendix
Table 7: Task-definition fields. Reward weights, optimizer settings, and network architecture are separate from this schema. Stored legacy success thresholds should not be confused with the paper’s reported 1 mm completion metric.
Entry
Absolute request
Policy entry
Interpretation
1
xk⋆
xk⋆−pt,x
Horizontal displacement, m
2
yk⋆
yk⋆−pt,y
Horizontal displacement, m
3
hk⋆
hk⋆−ht
Height error relative to the sheet, m
4
fk⋆
fk⋆
Requested force, N; not force error or measured contact force
5
tk⋆
tk⋆−t
Due-time offset, s; negative when the nominal due time has passed
Appendix
Table 8: Semantics of one five-dimensional trajectory/cursor sample. Normalization is applied after assembling the policy vector.
Figure 6: Matched episodes across IK, BC, and final RL (top), and their evolution from BC initialization through RL (bottom). Blue is deposited ink, gray is the target, and red marks target points not covered within 1 mm. Numbers report completion ∣ ink precision. The high-frequency deviations in IK and BC leave off-target ink; RL produces smoother, more target-aligned strokes.
Figure 7: Fifty-seed evaluations during final RL. The initial point is BC and dashed lines indicate IK on the same seeds. Snapshot timing makes the horizontal axis approximate by at most one rollout. All checkpoints come from one training run.
Figure 8: Twelve-seed task monitoring and training return/episode length during final RL. This monitoring set differs from the 50-seed aggregate evaluation and is descriptive rather than a repeated-training statistic.
Dexterous in-hand manipulation of a grasped object with an anthropomorphic hand is an unsolved frontier for robot dexterity. The contact-richness and highly dynamic nature of object-hand interactions tend to require extensive modeling or data-collection efforts for learning-based approaches. Modern simulators used for reinforcement learning (RL) cannot fully replicate the required contact complexity, while collecting dexterous demonstrations for imitation learning (IL) remains an open problem. In this research, we present an embodied control approach based on real-time task Jacobian estimation of the combined hand and object system on the physical robot. Using only the CPU on a laptop, the proposed controller begins in-hand pen writing after approximately 18 s of initialization and continues to adapt online, without an analytic hand--object kinematic/contact model, simulation training, or precollected task demonstrations. We demonstrate that the same estimator/controller formulation works on three anthropomorphic robotic hand systems (one physical, two simulated) to show human-like, in-hand articulation of a grasped pen by an embodiment-independent formulation. Sub-millimeter in-plane precision (mean 0.6 mm across runs) is achieved across letters and shapes written in the air and on paper on a physical robot. To our knowledge, this is the first demonstration of an anthropomorphic hand writing arbitrary single-stroke trajectories with a grasped pen through purely in-hand motion, and it showcases an alternative to compute- and data-heavy approaches such as RL and IL for achieving dexterous manipulation through computationally simple and data-efficient algorithms.
Kai Stewart, Yasunori Toshimitsu, Robert K. Katzschmann
Soft Robotics Lab, D-MAVT, ETH Zurich, 8092 Zurich, Switzerland
Human demonstrations offer a scalable data source for dexterous manipulation, but transferring them to robot actions remains challenging due to the embodiment gap. Today's retargeting is mostly kinematic, yet manipulation is decided by force, which governs how the hand interacts with the object and how the object moves. In this paper, we present ReForce, a Force-aware Retargeting method that turns human motion and forces into robot actions that reproduce the intended contact. ReForce predicts a residual on the kinematically retargeted action to reach the desired force, using a general force tracker trained on large-scale simulation interactions. It supports both online force-aware teleoperation and offline data translation. In simulation and on real hardware, ReForce achieves lower force-tracking error and stronger multi-finger contact engagement on contact-rich tasks such as paper-cup grasping and tongs manipulation.
A dexterous hand can change an object's pose without setting it down. This survey covers 113 learning-based methods that attempt it, the hands they run on, the simulators they train in, and how they report results. The field has largely converged on one recipe: reinforcement learning on privileged state in a GPU simulator, distilled into a vision-only student for transfer to hardware. Almost half the corpus trains that way, nearly all of it in one simulator, whose own successors are barely used. A second and faster-growing branch learns from teleoperated demonstration instead, through vision-language-action models, diffusion policies, and systems whose contribution is the collection rig itself. Published work has tripled since 2022. Grasping remains the largest task family and bimanual coordination the fastest-rising, and almost every experiment runs on one of four hands. What the survey adds comes from reading released code and vendor specifications beside the papers. Eight hands that can be bought or built from published designs appear in no method paper here. Where the depth a hand sinks into the object it holds is reported at all, it is not measured in the physics that produced the motion. Section VII proposes a protocol for the unreported quantities; the body gives the counts behind every claim above.