⚡ electrical engineering

Uncertainty-Aware Haptic Signal Estimation for Reliable and Resource Efficient Tactile Internet

The paper introduces the Agile AI-empowered Haptic (A2HAP) framework, which utilizes a novel probabilistic neural network and hierarchical gating to dynamically balance confidence and reliability, thereby significantly reducing network congestion and increasing user capacity for the Tactile Internet without compromising teleoperation stability.

Original authors: Georgios Kokkinis, Alexandros Iosifidis, Qi Zhang

Published 2026-06-16

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Georgios Kokkinis, Alexandros Iosifidis, Qi Zhang

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are playing a high-stakes video game where you control a robot arm on the other side of the world. To make it feel real, the robot needs to send you a "push" or "pull" signal (haptic feedback) hundreds of times every second. If the signal is even a tiny bit late or gets lost, the robot might jerk around, or you might feel like you're touching thin air.

The problem is that sending these signals so fast creates a massive traffic jam on the internet, especially when many people are doing it at once. It's like trying to shout a message to a friend in a crowded stadium; if everyone shouts at once, no one can hear anything.

This paper introduces a new system called A2HAP (Agile AI-empowered Haptic Communication) to solve this traffic jam without losing the "feel" of the robot. Here is how it works, using simple analogies:

1. The Problem: The "Shouting" Crowd

Currently, to keep the robot feeling smooth, systems send a signal every millisecond, regardless of whether anything interesting is happening.

The Old Way: Imagine a messenger running back and forth between you and the robot, shouting "I'm pushing!" "I'm pushing!" "I'm pushing!" even when the robot is just sitting still. This wastes energy and clogs the road.
The Limit: If too many people try to shout at once, the network crashes, and the robot stops working safely.

2. The Solution: A "Smart" Messenger with a Confidence Meter

The A2HAP system uses a special AI brain called VarxHAP. Think of this AI as a super-smart messenger who doesn't just repeat what happened; it predicts what will happen next.

The Prediction: Instead of waiting for the robot to tell the AI what it's doing, the AI watches the pattern and guesses, "Okay, the robot is moving left, so the next signal will probably be a push to the left."
The Confidence Meter (Uncertainty): This is the magic part. The AI doesn't just guess; it also has a "confidence meter."
- High Confidence: If the AI is 99% sure of its guess, it says, "I don't need to shout to the network! I'll just tell the robot what I think it should do." This saves a huge amount of space on the internet road.
- Low Confidence: If the robot does something sudden or weird (like hitting a wall), the AI's confidence meter drops. It says, "I'm not sure what's happening! I need to shout the real signal immediately so we don't get it wrong."

3. The Traffic Cop: The Error-Resilient Controller

The system has a smart traffic cop (the Controller) that watches the internet road.

When the road is empty: The traffic cop tells the AI, "You can be a bit more relaxed. If you're 90% sure, go ahead and skip sending the signal." This saves even more space.
When the road is jammed: If too many people are trying to send signals, the traffic cop gets strict. It tells the AI, "Stop guessing! Only send a signal if you are 100% sure, or send the real data immediately." This prevents the network from crashing.

4. The Results: Smoother Traffic, More People

The researchers tested this system in a simulation with 60 people trying to control robots at the same time.

Less Traffic: By letting the AI "skip" sending signals when it was confident, the system reduced the number of messages sent by up to 45% during busy times.
More Space: Because the road was less crowded, the system could handle 20% more users than previous methods without breaking the connection.
Safety: Even with fewer messages, the system kept the connection reliable (99.999% reliability), meaning the robot never lost control or felt "glitchy."

Summary

Think of the Tactile Internet as a busy highway. Old systems made every car honk its horn constantly, causing a traffic jam. The A2HAP system gives the drivers (the AI) a map and a confidence meter. If they know the road ahead is clear, they don't honk. If they see a hazard, they honk loudly. This keeps the highway flowing smoothly, allowing more cars to travel safely at the same time.

Technical Summary: Uncertainty-Aware Haptic Signal Estimation for Reliable and Resource Efficient Tactile Internet

Problem Statement

The Tactile Internet (TI) aims to enable real-time remote haptic interaction, requiring closed-loop control systems with sub-millisecond latency and extreme reliability (exceeding 99.999%). A primary bottleneck in scaling these systems is the high sampling rate (typically 1 kHz) necessary for perceptual transparency. In multi-user wireless environments, transmitting at such rates causes severe network congestion, interference, and deadline violations.

While traditional methods like Perceptual Deadband (PD) filtering successfully suppress packets during low-motion activity, they fail to prevent traffic spikes during high-motion bursts. Furthermore, existing predictive models (e.g., LSTM, Gaussian processes) often operate independently of the MAC layer, lacking mechanisms to translate model confidence into real-time packet suppression decisions. Additionally, current approaches struggle to balance the trade-off between reducing transmission rates and maintaining the ultra-reliability required for stable teleoperation under varying network conditions.

Methodology: The A2HAP Framework

To address the disconnect between predictive AI and communication constraints, the authors propose the Agile AI-empowered Haptic (A2HAP) framework. This system integrates a probabilistic neural network with an error-resilient controller to dynamically adapt transmission thresholds.

1. Core Components

VarxHAP (Probabilistic Estimator): A novel neural network that outputs both a force estimate and a calibrated uncertainty metric. It operates on both the operator and teleoperator sides, utilizing a dual-branch Gated Recurrent Unit (GRU) architecture with cross-attention to fuse historical force feedback with current operator commands (position and velocity).
- Step-Aware Embedding: To handle high-frequency haptics, the model incorporates a step-aware embedding that explicitly maps estimation steps to a high-dimensional space, addressing the inability of standard RNNs to track rollout depths.
- Composite Uncertainty Loss: The model is trained using a composite loss function combining point prediction error (MSE), a Gaussian Negative Log-Likelihood (NLL) term to penalize low uncertainty during high-error events, and a calibration term to force the predicted variance to match observed squared residuals.
Error-Resilient Controller: A closed-loop mechanism that dynamically scales transmission thresholds based on real-time network congestion and model uncertainty.
- Uncertainty-Aware Suppression: The system employs a hierarchical gating logic. A transmission is requested only if the perceptual change exceeds the Just Noticeable Difference (JND) threshold AND the model's uncertainty exceeds a dynamic threshold $\tau_k(m)$ .
- Adaptive Thresholding: The threshold $\tau_k(m)$ is exponentially decayed based on the number of consecutive estimations to prevent drift.
- PI Control Logic: A Proportional-Integral (PI) controller adjusts the base uncertainty ( $\sigma_b$ ) and decay rate ( $\lambda$ ) parameters. It distinguishes between errors caused by estimation drift and those caused by network congestion, adjusting thresholds to either force more frequent transmissions (to refresh the estimator) or suppress more packets (to alleviate congestion).

2. Signal Reconstruction

At the receiver, the haptic signal is reconstructed based on transmission success:

Success: The true force is received.
Perceptual Silence ( $f_{JND}=0$ ): The last known reference is replayed.
Model Confidence ( $f_{AI}=0$ ) or Error: The receiver uses the VarxHAP point prediction ( $\mu_{k,t}$ ) as a fallback recovery mechanism.

Key Contributions

The paper outlines three primary contributions:

Uncertainty-Aware Haptic Estimator (VarxHAP): A probabilistic model providing force estimates alongside calibrated uncertainty, enabling both packet suppression and signal recovery.
Error-Resilient Controller: A closed-loop mechanism that dynamically scales transmission thresholds based on model uncertainty. It ensures the strict 99.999% reliability target is maintained by determining precisely when to suppress physical packets in favor of local model inference.
AI-Empowered Framework (A2HAP): An agile framework that proactively replaces physical transmissions with confident model predictions rather than relying solely on reactive perceptual deadbands.

Simulation Results

The framework was evaluated in a simulated environment with 60 users, variable Packet Error Rates (PER), and round-robin Resource Block (RB) allocation using real-world 1 kHz haptic traces.

Resource Efficiency: A2HAP reduced Resource Block consumption by 25% on average compared to baseline methods. During peak traffic bursts, packet rates were suppressed by up to 45%.
User Capacity: The framework supported a 20% increase in user capacity compared to state-of-the-art methods (accommodating 60 users vs. 50 for the baseline) while strictly maintaining the 99.999% reliability target.
Reliability Stability: Unlike baselines that suffer rapid performance degradation under congestion, A2HAP demonstrated graceful degradation. The error-resilient controller successfully balanced congestion-induced errors against estimation errors, maintaining high reliability even under adverse channel conditions (PER up to $10^{-3}$ ).
Uncertainty Calibration: The VarxHAP model demonstrated high accuracy in force estimation while correctly assigning high uncertainty to volatile regions (sharp force changes), validating its utility as a gating mechanism.

Significance and Claims

The paper claims that A2HAP represents a paradigm shift from reactive to predictive, uncertainty-aware haptic communication. By bridging the gap between AI-native estimation and the stringent requirements of Ultra-Reliable Low-Latency Communications (URLLC), the framework provides a calibrated confidence metric that serves as a robust control variable for network resource management.

The authors assert that integrating uncertainty directly into the transmission loop is a fundamental enabler for the Tactile Internet, allowing for significant resource efficiency gains during low traffic and reliability stabilization during high traffic. The work demonstrates that AI-driven packet suppression, when coupled with an error-resilient controller, can meet the ultra-reliability demands of teleoperation without compromising the user experience or network stability.