TATIC: Task-Aware Temporal Intent Inference

• TATIC is a task-aware temporal learning framework that converts brief physical corrections into discrete task-level intent and continuous parameters for human-robot collaboration.
• It leverages torque-based contact estimation, task-aligned feature canonicalization, and a causal Temporal Convolutional Network to robustly infer operator intent under low-latency conditions.
• Empirical validation shows high intent recognition accuracy and responsive motion adaptation, while limitations include a fixed intent vocabulary and potential ambiguity in physical corrections.

to=arxiv_search.search 下载彩神争霸াত্র json {"^^^^^^^^5query5 arXiv human robot collaboration physical corrections intent inference 20265^^^^^^^^", "^^^^^^^^5max_results5 5} to=arxiv_search.search 全民彩票天天送json {"^^^^^^^^5query5 Temporal Learning for Human Intent Inference from Physical Corrections in Human-Robot Collaboration^^^^^^^^5\5 "^^^^^^^^5max_results5 5} to=arxiv_search.search สล็อตโjson {"^^^^^^^^5query5 TacSIm Tactical Style Imitation football arXiv", "^^^^^^^^5max_results5 ^^^^5TATIC arXiv human robot collaboration physical corrections intent inference 20265query5^^^^} TATIC most directly denotes Task-Aware Temporal Learning for Human Intent Inference from Physical Corrections in Human-Robot Collaboration, a framework for decoding brief physical corrections in human-robot collaboration into both discrete task-level intent and continuous motion-level parameters. It combines torque-based contact force estimation, task-aligned feature canonicalization, a causal task-aware Temporal Convolutional Network (TCN), and an intent-driven adaptation layer that maps inferred semantics into path deformation, operation-space inflation, speed scaling, goal switching, or software-level hold. The formulation is motivated by contact-rich collaborative settings in which operators convey motion-level adjustments through short physical interactions, while the robot must recover task-level semantics under dynamic constraints and low-latency requirements (&&&^^^^5query5^^^^&&&).

^^^^5TATIC arXiv human robot collaboration physical corrections intent inference 20265^^^^. Scope, intent vocabulary, and problem setting

TATIC addresses a specific gap in human-robot collaboration: existing foundation-model-based approaches primarily rely on vision and language inputs and lack mechanisms to interpret physical feedback, while traditional physical human-robot interaction methods use corrections for trajectory guidance but struggle to infer task-level semantics from brief contacts (&&&^^^^5query5^^^^&&&). The core problem is therefore not only motion adaptation, but the joint recovery of task-level intent and operator-conditioned continuous parameters from contact segments whose durations were brief, specifically ^^^^5query5^^^^.5^^^^5query5 arXiv human robot collaboration physical corrections intent inference 20265^^^^.98 s.

The discrete intent vocabulary is fixed to

PRESERVED_PLACEHOLDER_^^^^5query5^^^^

TATIC jointly estimates continuous quantities conditioned on the inferred operator: for Guide, a direction PRESERVED_PLACEHOLDER_^^^^5TATIC arXiv human robot collaboration physical corrections intent inference 20265^^^^ and magnitude PRESERVED_PLACEHOLDER_^^^^5max_results5^^^^; for Yield, a safety-margin radius PRESERVED_PLACEHOLDER_^^^^5query5^^^^; for Slow, a speed scaling PRESERVED_PLACEHOLDER_^^^^5\5^^^^; and for Switch, a target index $k \in \{0,1\}$ (&&&^^^^5query5^^^^&&&). This design separates semantic intent from low-level execution variables while keeping the latter explicitly actionable.

Several challenges are explicit in the formulation. First, contact segments are short, so inference must be causal and low-latency. Second, task constraints such as task-centric safety zones and planner ambiguity vary online. Third, workspace reconfigurations change global pose, so a representation tied to world coordinates degrades generalization. TATIC addresses this by constructing a task-aligned canonical frame and learning over canonicalized temporal feature streams rather than raw global geometry. A plausible implication is that the framework is best understood as a task-conditioned temporal inference system rather than a generic force-to-action mapper.

^^^^5max_results5^^^^. Torque-based contact estimation and contact-segment formation

The sensing pipeline begins with measured joint torques $\tau_{\mathrm{meas}}[k]$, together with proprioceptive variables $q[k]$, $\dot q[k]$, and $\ddot q[k]$. The dynamics model is

PRESERVED_PLACEHOLDER_^^^^5TATIC arXiv human robot collaboration physical corrections intent inference 20265query5^^^^

and the formulation used in TATIC collects Coriolis/centripetal, gravity, and friction into PRESERVED_PLACEHOLDER_^^^^5TATIC arXiv human robot collaboration physical corrections intent inference 20265TATIC arXiv human robot collaboration physical corrections intent inference 20265^^^^, giving

PRESERVED_PLACEHOLDER_^^^^5TATIC arXiv human robot collaboration physical corrections intent inference 20265max_results5^^^^

External torque from a contact at PRESERVED_PLACEHOLDER_^^^^5TATIC arXiv human robot collaboration physical corrections intent inference 20265query5^^^^ is modeled as

PRESERVED_PLACEHOLDER_^^^^5TATIC arXiv human robot collaboration physical corrections intent inference 20265\5^^^^

The residual torque estimate is then

PRESERVED_PLACEHOLDER_^^^^5TATIC arXiv human robot collaboration physical corrections intent inference 202655^^^^

and the contact detection statistic is

PRESERVED_PLACEHOLDER_^^^^5TATIC arXiv human robot collaboration physical corrections intent inference 20265^^^^6

combined with EWMA filtering, a threshold PRESERVED_PLACEHOLDER_^^^^5TATIC arXiv human robot collaboration physical corrections intent inference 20265^^^^7, and persistence logic PRESERVED_PLACEHOLDER_^^^^5TATIC arXiv human robot collaboration physical corrections intent inference 20265^^^^8 for onset and offset detection (&&&^^^^5query5^^^^&&&).

Once contact is detected, TATIC localizes the contacted link coarsely by thresholding active joint residuals and choosing the highest active joint satisfying a persistence condition. Contact position is parameterized by a normalized scalar PRESERVED_PLACEHOLDER_^^^^5TATIC arXiv human robot collaboration physical corrections intent inference 20265^^^^9 along the contacted link centerline. Over a short window of samples PRESERVED_PLACEHOLDER_^^^^5max_results5query5^^^^, the residual model is

PRESERVED_PLACEHOLDER_^^^^5max_results5TATIC arXiv human robot collaboration physical corrections intent inference 20265^^^^

and the contact state is estimated by solving

PRESERVED_PLACEHOLDER_^^^^5max_results5max_results5^^^^

For fixed PRESERVED_PLACEHOLDER_^^^^5max_results5query5^^^^, PRESERVED_PLACEHOLDER_^^^^5max_results5\5^^^^ is obtained by Tikhonov-regularized least-squares and projected onto the ball PRESERVED_PLACEHOLDER_^^^^5max_results55^^^^; PRESERVED_PLACEHOLDER_^^^^5max_results5^^^^6 is searched on a uniform grid and refined via Brent search (&&&^^^^5query5^^^^&&&).

This stage is central because all downstream semantics are conditioned on the estimated contact geometry and force. The framework therefore does not treat physical correction as an abstract symbol stream; it explicitly reconstructs a physically grounded interaction state from residual dynamics.

^^^^5query5^^^^. Task-aligned feature canonicalization

A defining element of TATIC is task-aligned feature canonicalization, introduced to improve robustness under layout reconfiguration. Rather than learning directly in the world frame, the method constructs a local canonical frame PRESERVED_PLACEHOLDER_^^^^5max_results5^^^^7 aligned with the planner-provided reference velocity PRESERVED_PLACEHOLDER_^^^^5max_results5^^^^8 and the world vertical axis PRESERVED_PLACEHOLDER_^^^^5max_results5^^^^9 (&&&^^^^5query5^^^^&&&).

The forward axis is

PRESERVED_PLACEHOLDER_^^^^5query5query5^^^^

otherwise PRESERVED_PLACEHOLDER_^^^^5query5TATIC arXiv human robot collaboration physical corrections intent inference 20265^^^^ is reused. An auxiliary axis is selected as

PRESERVED_PLACEHOLDER_^^^^5query5max_results5^^^^

otherwise PRESERVED_PLACEHOLDER_^^^^5query5query5^^^^. The remaining basis vectors are

PRESERVED_PLACEHOLDER_^^^^5query5\5^^^^

yielding the raw frame PRESERVED_PLACEHOLDER_^^^^5query55^^^^. Temporal smoothing on PRESERVED_PLACEHOLDER_^^^^5query5^^^^6 is performed via

PRESERVED_PLACEHOLDER_^^^^5query5^^^^7

Positions and vectors are then projected into the task frame. For positions,

PRESERVED_PLACEHOLDER_^^^^5query5^^^^8

and for general vectors,

PRESERVED_PLACEHOLDER_^^^^5query5^^^^9

The canonicalized feature set includes kinematics, workspace safety, goal alignment, localization, and planner context (&&&^^^^5query5^^^^&&&). Representative terms are the reference-speed magnitude PRESERVED_PLACEHOLDER_^^^^5\5query5^^^^; local unit force direction

PRESERVED_PLACEHOLDER_^^^^5\5TATIC arXiv human robot collaboration physical corrections intent inference 20265^^^^

workspace clearance

PRESERVED_PLACEHOLDER_^^^^5\5max_results5^^^^

escape direction

PRESERVED_PLACEHOLDER_^^^^5\5query5^^^^

and goal-alignment features PRESERVED_PLACEHOLDER_^^^^5\5\5^^^^, PRESERVED_PLACEHOLDER_^^^^5\55^^^^, and PRESERVED_PLACEHOLDER_^^^^5\5^^^^6, which encode target preference for Switch.

The full feature vector is assembled as

PRESERVED_PLACEHOLDER_^^^^5\5^^^^7

Empirically, this representation shift is consequential: on the canonicalization ablation, the canonical frame achieved ^^^^5query5^^^^.9^^^^5query5\5 / ^^^^5query5^^^^.87^^^^5TATIC arXiv human robot collaboration physical corrections intent inference 20265^^^^ intent-recognition F^^^^5TATIC arXiv human robot collaboration physical corrections intent inference 20265^^^^ on ID test / OOD-Reconfig, whereas the world frame yielded ^^^^5query5^^^^.889 / ^^^^5query5^^^^.6^^^^5TATIC arXiv human robot collaboration physical corrections intent inference 20265\5^^^^, and world + SE(^^^^5max_results5^^^^) aug yielded ^^^^5query5^^^^.89^^^^5TATIC arXiv human robot collaboration physical corrections intent inference 20265^^^^ / ^^^^5query5^^^^.75^^^^5query5 (&&&^^^^5query5^^^^&&&). This suggests that the primary benefit is not simple augmentation, but the task-relative restructuring of geometry itself.

^^^^5\5^^^^. Task-aware Temporal Convolutional Network

TATIC feeds the causal temporal window

PRESERVED_PLACEHOLDER_^^^^5\5^^^^8

into a task-aware TCN. Inputs are standardized using training statistics, and independent dropout is applied to the localization subvector PRESERVED_PLACEHOLDER_^^^^5\5^^^^9. A compact MLP encoder $k \in \{0,1\}$^^^^5query5^^^^ maps each $k \in \{0,1\}$^^^^5TATIC arXiv human robot collaboration physical corrections intent inference 20265^^^^ to an embedding $k \in \{0,1\}$^^^^5max_results5^^^^ with channel width $k \in \{0,1\}$^^^^5query5^^^^. Temporal processing uses $k \in \{0,1\}$^^^^5\5^^^^ residual blocks with kernel size $k \in \{0,1\}$5 and exponentially increasing dilations $k \in \{0,1\}$6, followed by last-step pooling to produce the latent state $k \in \{0,1\}$7 (&&&^^^^5query5^^^^&&&).

The receptive field is

$k \in \{0,1\}$8

For doubling dilations $k \in \{0,1\}$9, $\tau_{\mathrm{meas}}[k]$^^^^5query5^^^^, giving

$\tau_{\mathrm{meas}}[k]$^^^^5TATIC arXiv human robot collaboration physical corrections intent inference 20265^^^^

steps, which the paper states covers the full window. The network then branches into decoupled classification and regression heads,

$\tau_{\mathrm{meas}}[k]$^^^^5max_results5^^^^

The classification branch outputs logits over the five operators and the two switch targets; the regression branch predicts $\tau_{\mathrm{meas}}[k]$^^^^5query5^^^^, $\tau_{\mathrm{meas}}[k]$^^^^5\5^^^^, $\tau_{\mathrm{meas}}[k]$5, and $\tau_{\mathrm{meas}}[k]$6, with domain clamping applied after prediction (&&&^^^^5query5^^^^&&&).

Training uses focal cross-entropy for operator classification, cross-entropy for target selection, masked cosine loss for Guide direction, masked MSE for scalar regressands, and homoscedastic uncertainty weighting across tasks: $\tau_{\mathrm{meas}}[k]$7 where $\tau_{\mathrm{meas}}[k]$8. The direction term is

$\tau_{\mathrm{meas}}[k]$9

Optimization uses AdamW with a cosine learning-rate schedule, and the full model has approximately ^^^^5query5^^^^.6^^^^5query5 trainable parameters (&&&^^^^5query5^^^^&&&).

The architecture is explicitly causal and therefore designed for online deployment. In contrast to sequence models that depend on bidirectional context, TATIC constrains itself to the information available within the active contact segment.

5. Intent-conditioned adaptation and empirical validation

The output of TATIC is not merely a label; it is a control-relevant semantic state that drives distinct adaptation modes. For Guide, the executed path is

$q[k]$^^^^5query5^^^^

with semantic displacement

$q[k]$^^^^5TATIC arXiv human robot collaboration physical corrections intent inference 20265^^^^

The incremental displacement is injected over a bounded horizon using a bump function

$q[k]$^^^^5max_results5^^^^

ensuring endpoint preservation (&&&^^^^5query5^^^^&&&).

For Yield, the predicted safety radius is mapped into an inflated operation space via Minkowski sum,

$q[k]$^^^^5query5^^^^

For Slow, the reference velocity is scaled as

$q[k]$^^^^5\5^^^^

with a first-order low-pass filter enforcing acceleration limits. For Switch, the chosen mission goal is

$q[k]$5

and the planner replans accordingly. Stop triggers a software-level hold at the current pose (&&&^^^^5query5^^^^&&&).

The evaluation dataset comprises a 7-DoF manipulator, 5^^^^5query5query5^^^^ in-distribution episodes across ^^^^5TATIC arXiv human robot collaboration physical corrections intent inference 20265query5query5^^^^ trajectories, and ^^^^5max_results55query5^^^^ out-of-distribution episodes across 5^^^^5query5^^^^ reconfigured trajectories. The ID split is 7^^^^5query5^^^^/^^^^5TATIC arXiv human robot collaboration physical corrections intent inference 202655^^^^/^^^^5TATIC arXiv human robot collaboration physical corrections intent inference 202655^^^^ at trajectory level, with OOD-Reconfig reserved for OOD evaluation. On the test set, TATIC achieved Macro-F^^^^5TATIC arXiv human robot collaboration physical corrections intent inference 20265^^^^ = ^^^^5query5^^^^.9^^^^5query5\5, ECE = ^^^^5query5^^^^.^^^^5query5\5TATIC arXiv human robot collaboration physical corrections intent inference 20265^^^^, and 95% bootstrap CI [^^^^5query5^^^^.887, ^^^^5query5^^^^.9^^^^5max_results5max_results5 for intent recognition. Per-class F^^^^5TATIC arXiv human robot collaboration physical corrections intent inference 20265^^^^ scores were Guide ^^^^5query5^^^^.86^^^^5max_results5, Yield ^^^^5query5^^^^.879, Slow ^^^^5query5^^^^.9^^^^5max_results5, Stop ^^^^5query5^^^^.9^^^^5query5TATIC arXiv human robot collaboration physical corrections intent inference 20265^^^^, and Switch ^^^^5query5^^^^.95^^^^5TATIC arXiv human robot collaboration physical corrections intent inference 20265^^^^ (&&&^^^^5query5^^^^&&&).

Regression performance was also reported. End-to-end results were Guide direction cosine similarity ^^^^5query5^^^^.89^^^^5TATIC arXiv human robot collaboration physical corrections intent inference 20265^^^^, Guide magnitude RMSE ^^^^5query5^^^^.^^^^5query5, Slow scaling RMSE ^^^^5query5^^^^.^^^^5query5, Yield radius RMSE ^^^^5query5^^^^.^^^^5TATIC arXiv human robot collaboration physical corrections intent inference 20265query5^^^^8, and Switch target F^^^^5TATIC arXiv human robot collaboration physical corrections intent inference 20265^^^^ ^^^^5query5^^^^.9^^^^5max_results5TATIC arXiv human robot collaboration physical corrections intent inference 20265^^^^. With ground-truth operator filtering, the corresponding values were ^^^^5query5^^^^.9^^^^5TATIC arXiv human robot collaboration physical corrections intent inference 20265^^^^6, ^^^^5query5^^^^.^^^^5query5query5\5, ^^^^5query5^^^^.^^^^5query5max_results5TATIC arXiv human robot collaboration physical corrections intent inference 20265^^^^, ^^^^5query5^^^^.^^^^5query5, and ^^^^5query5^^^^.9^^^^5query5 (&&&^^^^5query5^^^^&&&).

Ablations isolate the role of feature groups. Using kinematics only gave ^^^^5query5^^^^.58^^^^5query5 / ^^^^5query5^^^^.5^^^^5max_results5\5 / ^^^^5query5^^^^.7^^^^5\5TATIC arXiv human robot collaboration physical corrections intent inference 20265^^^^ / ^^^^5query5^^^^.^^^^5TATIC arXiv human robot collaboration physical corrections intent inference 20265TATIC TacSIm Tactical Style Imitation football arXiv5max_results5^^^^ for operator Macro-F^^^^5TATIC arXiv human robot collaboration physical corrections intent inference 20265^^^^, target F^^^^5TATIC arXiv human robot collaboration physical corrections intent inference 20265^^^^, Guide direction cosine, and magnitude RMSE; adding workspace features improved this to ^^^^5query5^^^^.7^^^^5TATIC arXiv human robot collaboration physical corrections intent inference 20265max_results5^^^^ / ^^^^5query5^^^^.557 / ^^^^5query5^^^^.8^^^^5query5query5 / ^^^^5query5^^^^.^^^^5TATIC arXiv human robot collaboration physical corrections intent inference 20265\5^^^^9; adding alignment features gave ^^^^5query5^^^^.8^^^^5\5TATIC arXiv human robot collaboration physical corrections intent inference 20265^^^^ / ^^^^5query5^^^^.89^^^^5TATIC arXiv human robot collaboration physical corrections intent inference 20265^^^^ / ^^^^5query5^^^^.87^^^^5max_results5 / ^^^^5query5^^^^.^^^^5TATIC arXiv human robot collaboration physical corrections intent inference 20265TATIC arXiv human robot collaboration physical corrections intent inference 20265max_results5^^^^; and the full TATIC feature set reached ^^^^5query5^^^^.9^^^^5query5\5 / ^^^^5query5^^^^.9^^^^5query5 / ^^^^5query5^^^^.9^^^^5TATIC arXiv human robot collaboration physical corrections intent inference 20265^^^^6 / ^^^^5query5^^^^.^^^^5query5query5\5 (&&&^^^^5query5^^^^&&&). This indicates that semantic intent recognition in the framework is strongly dependent on task-relative geometric structure rather than kinematics alone.

Hardware validation was conducted in collaborative desktop disassembly, using a 7-DoF manipulator, side-mounted and wrist-mounted RGB-D cameras for target coordinates, torque-based contact estimation for physical corrections, a constraint-based task planner from component connectivity and prerequisites, and cuRobo for nominal references. Reported examples include lateral pushes decoded as Guide, gentle pushes near a boundary decoded as Yield, taps decoded as Slow, directional corrections decoded as Switch, and assertive holds decoded as Stop (&&&^^^^5query5^^^^&&&). The reported outcome was reliable semantic inference and responsive motion adaptation in hardware.

6. Limitations, failure modes, and nomenclature

The framework’s stated limitations are explicit. It uses a predefined intent vocabulary and does not yet model personalization across operators. Reported failure modes include ambiguous corrections, such as Guide versus Yield confusion for lateral pushes, and contact estimation drift under modeling errors or multi-contact scenarios. Planned extensions include adaptive personalization, expanded intent semantics, larger datasets, and multimodal fusion across vision, language, and force (&&&^^^^5query5^^^^&&&).

The acronym itself is not stable across adjacent literatures. In football analytics, TATIC is used as shorthand for Tactical Style Imitation, where the objective is to reproduce a team’s spatio-temporal tactical organization rather than infer intent from physical corrections; TacSIm operationalizes that usage as a benchmark grounded in broadcast footage and virtual-environment evaluation (&&&^^^^5TATIC arXiv human robot collaboration physical corrections intent inference 20265^^^^8&&&). By contrast, the teamwork-intervention framework ATTIC/TIC is a different concept entirely: it learns a generative model of team behavior with BTIL and generates task-time interventions for coordination improvement, and the authors explicitly note that “TATIC” does not appear in that paper and is only a naming variant encountered elsewhere (&&&^^^^5TATIC arXiv human robot collaboration physical corrections intent inference 20265^^^^9&&&).

Within human-robot collaboration, however, TATIC is most precisely identified with the formulation in which torque residuals, contact geometry, task-frame canonicalization, and a causal TCN are combined to infer discrete operators and continuous task parameters from brief physical corrections. A plausible implication is that its main contribution lies in converting physical contact from a low-level compliance signal into a structured semantic control channel for online collaborative manipulation.