Source-linked AI summary
Opportunistic Data Offloading for Robotic Operations in Dynamically Varying Channel Environments
Heeirthan Shanthan, Winston Hurst, Yasamin Mostofi
TL;DR
Dynamic obstacles can change mmWave channel conditions abruptly, complicating joint motion and communication planning for AVs that must transmit onboard data promptly. The paper proposes predictive NMPC with a CBF safety filter; simulations report up to 37.3% lower total energy than baseline strategies.
Problem
Moving obstacles create abrupt mmWave channel variations that tightly couple AV motion and communication decisions, challenging timely, energy-efficient data offloading.
Source: §I · p. 1Method
A real-time NMPC framework jointly optimizes motion and communication while a CBF safety filter maintains obstacle safety margins.
Source: §I · p. 1, C. CBF Safety Filter · p. 4Results
Up to 37.3% lower total energy consumption than baseline strategies was reported in simulated environments.
Source: §VI · p. 6Takeaways & Limitations
The method reduces communication energy by orders of magnitude when NLOS regions are unavoidable in the simulated environments.
Source: §VI · p. 6
Abstract
from arXiv · showhide
This paper studies energy-efficient operation of autonomous vehicles (AVs) in dynamic environments with moving obstacles and while communicating over mmWave channels. The obstacles induce severe attenuation of the mmWave channel resulting in a highly dynamic communication environment. In this setting, we consider the problem of jointly optimizing motion and communication energy for an AV that safely navigates among dynamic obstacles toward a designated destination while ensuring timely transmission of onboard sensing or telemetry data over mmWave channels. We then seek a real-time methodology to compute energy-efficient trajectories in a setting where dynamic obstacles induce both safety constraints and time-varying mmWave blockage, leading to tightly coupled motion-communication trade-offs. We propose a nonlinear model predictive control (NMPC) framework that enables anticipative communication and motion decision-making and energy co-optimization, augmented with a control barrier function (CBF) to ensure safety. Extensive simulation results demonstrate the effectiveness of our approach, reducing total energy consumption by up to 37.3% compared to baseline strategies. Overall, our results demonstrate that the proposed NMPC-based framework significantly enhances energy efficiency and performance of AVs under dynamic, blockage-sensitive mmWave communication constraints.
I. INTRODUCTION
AVs need reliable communication while navigating dynamic environments, but moving obstacles can make mmWave links vary abruptly and couple route planning with communication decisions. The paper addresses this challenge through joint motion and communication optimization, with onboard buffer constraints shaping timely data offloading.
I. INTRODUCTION
- Moving obstacles can abruptly change mmWave links between LOS and NLOS, while prior co-optimization approaches typically assume static environments or communication models that do not capture mmWave blockage and directionality.This leaves motion and communication decisions tightly coupled in dynamic settings.Source: §I · p. 1
- The proposed real-time NMPC framework combines a CBF safety filter with predictive motion and communication decisions to reduce energy and prevent buffer overflow in simulations.The introduction reports total energy reductions of up to 37.3% versus baseline methods while completing the mission.Source: §I · p. 1
II. SYSTEM MODELING
- The AV accumulates sensor data in a finite onboard buffer at rate ψ and selects an offloading rate Rk while navigating around obstacles and maintaining a link to the base station.Obstacles can block the direct LOS link, increasing channel attenuation.Source: §II · p. 1, Figure 1 · p. 2, A. Onboard Buffer Dynamics · p. 2
- The buffer update accounts for incoming data and offloaded data, and its length must remain between zero and its maximum capacity, which also bounds data age.The maximum buffer length can be interpreted as a constraint on the maximum age of information.Source: A. Onboard Buffer Dynamics · p. 2
B. Communication and Channel Model
The communication model links transmission requirements to channel quality and blockage, while the vehicle model supplies the state and control variables used for prediction. The stated assumptions include accurate current-state measurements and use of the same vehicle model for prediction.
B. Communication and Channel Model
- The minimum transmission power for a target spectral efficiency is modeled from channel-to-noise ratio and a maximum acceptable bit error rate, with channel power determined by path loss and Rician fading.The model uses distinct LOS and NLOS parameters, with LOS yielding stronger CNR than NLOS.Source: B. Communication and Channel Model · p. 2
C. Dynamic Model
- The AV state comprises planar position, heading, and buffer length, while linear velocity, angular velocity, and offloading rate are control inputs.Acceleration bounds are enforced through rate constraints on successive control inputs.Source: C. Dynamic Model · p. 2
- The prediction framework assumes accurate current AV-state measurements and uses the same dynamic model for prediction.These are explicit modeling assumptions for the optimization.Source: C. Dynamic Model · p. 2
D. Dynamic Obstacles
The formulation accounts for dynamic obstacles whose future positions are predicted internally, although actual positions may differ. It jointly seeks low motion and communication energy while meeting destination, data-offloading, safety, and real-time requirements.
D. Dynamic Obstacles
- Current obstacle positions are assumed accurately measured, and future motion is predicted deterministically even though actual positions may deviate from those predictions.The paper considers prediction deviations in its later simulation section.Source: D. Dynamic Obstacles · p. 2
III. PROBLEM FORMULATION
- The optimization minimizes motion and communication energy while reaching the destination and offloading data promptly, subject to vehicle dynamics, obstacle avoidance, and buffer constraints.Transmission power is limited, and an additional hardware maximum-rate constraint applies when the channel-based rate bound exceeds physical capability.Source: §III · pp. 2–3
- The objective penalizes control effort, acceleration, and transmission power, with weights tuning the relative importance of motion, communication, and prompt arrival.The motion-cost weights exclude the transmission-rate component of the control input.Source: §III · p. 3
- Motion-efficient and communication-efficient choices can conflict, while nonconvex dynamics and obstacle constraints and the need for timely arrival complicate real-time solution.A communication-efficient route may seek high-CNR regions, whereas a motion-efficient route may favor the shortest feasible path.Source: §III · p. 3
IV. NMPC-CBF OPTIMIZATION FRAMEWORK
The proposed NMPC approximates the optimization with finite-horizon predictive control and tracks a precomputed reference path to promote timely progress. A CBF safety filter minimally adjusts the NMPC output when needed to maintain obstacle safety margins.
IV. NMPC-CBF OPTIMIZATION FRAMEWORK
- At each time step, NMPC optimizes motion and transmission rates over a finite prediction horizon and applies the first control input from the resulting sequence.The reference path is the shortest path between initial and final positions and serves as an NMPC tracking objective.Source: §IV · p. 3
- A CBF safety filter minimally modifies the NMPC output when necessary to keep the AV outside obstacle safety margins.The updated state and obstacle measurements feed the next optimization step.Source: §IV · p. 3
A. NMPC Formulation
The NMPC formulation optimizes a sequence of controls over a prediction horizon, balancing waypoint tracking, destination progress, and other objectives. It applies the first control and repeats the optimization using updated measurements.
A. NMPC Formulation
- Over horizon N, NMPC optimizes a control sequence using stage and terminal costs subject to the predicted system dynamics.The state prediction follows z_j+1 = f(z_j, u_j).Source: A. NMPC Formulation · p. 3
- Waypoint costs penalize deviation from the reference path, while goal costs allow communication-motivated detours toward higher CNR while maintaining progress to the destination.The relative cost weights balance motion, communication, and arrival-time objectives.Source: A. NMPC Formulation · pp. 3–4
- At each time step, the controller applies only the first computed input, then reoptimizes with updated measurements using PANOC for real-time solution.PANOC is described as computationally efficient and able to handle nonlinearities and nonconvexities.Source: A. NMPC Formulation · p. 4
B. Prediction Models
The prediction models use current state and obstacle measurements, deterministic predictions for future obstacle motion, and expected channel gain for tractability. Simulations evaluate performance with stochastic small-scale fading realizations.
B. Prediction Models
- The model assumes accurate current AV-state and obstacle-position measurements, then predicts future obstacle positions with a deterministic linear motion model.The simulations account for deviations of actual obstacle positions from that prediction.Source: B. Prediction Models · p. 4
- For channel prediction, path loss uses predicted AV–base-station distance and LOS/NLOS parameters, while expected small-scale fading simplifies to E[H] = HPL.The expected model depends only on path loss and the LOS/NLOS condition.Source: B. Prediction Models · p. 4
- Performance simulations use stochastic small-scale fading realizations, and the adopted mmWave channel parameters are experimentally validated.The passage notes that LOS/NLOS parameters can be learned from prior channel measurements.Source: B. Prediction Models · p. 4
C. CBF Safety Filter
Because NMPC alone may violate obstacle safety margins under real-time implementation and solver tolerances, the framework adds a CBF safety filter. The filter selects controls that render the safe set forward invariant while staying close to the NMPC input.
C. CBF Safety Filter
- NMPC provides no formal safety guarantee and may violate obstacle safety margins because of real-time implementation and solver tolerances.A CBF is incorporated as a safety filter to enforce safety.Source: C. CBF Safety Filter · p. 4
- The CBF control objective is to render the set A = {z | h(z) ≥ 0} forward invariant while keeping the control close to the nominal NMPC input.The control is computed by solving a quadratic program.Source: C. CBF Safety Filter · p. 4
- For circular dynamic obstacles, h(z) is defined from the AV’s position relative to the obstacle center and its safety margin.The obstacle center is (x0, y0), and rsafe defines the safety margin.Source: C. CBF Safety Filter · p. 4
V. SIMULATION RESULTS
Simulations indicate that the proposed method lowers total energy, particularly when NLOS regions are unavoidable. The comparison includes motion-only and fixed-rate communication-aware baselines, while the proposed method adapts its rate and route to data needs.
V. SIMULATION RESULTS
- The proposed method significantly lowers total energy compared with baselines, particularly when NLOS regions are unavoidable.The supplied passage gives no numerical reduction for these simulation results.Source: §V · p. 4
- As incoming data rate ψ changes, the method adapts communication scheduling and route planning to communication needs.The simulations assess trajectories generated under different incoming data rates.Source: §V · p. 4
- The motion baseline ignores communication costs and constraints, while the communication baseline includes them but fixes transmission at R = ψ, requiring movement toward strong-CNR regions to reduce communication energy.The communication baseline therefore lacks adaptive rate optimization.Source: §V · p. 4
A. Simulation Setup
The simulations use a 10 m × 10 m workspace with moving obstacles and a mmWave channel, alongside fixed QoS, buffer, and robot parameters. NMPC horizons and motion-energy weights are specified for the two scenarios.
A. Simulation Setup
- The simulations use a 10 m × 10 m workspace, with obstacle motion modeled linearly plus additive Gaussian noise.The QoS target is pBER = 10^-6, channel bandwidth is 100 MHz, and receiver noise power is −85 dBm.Source: A. Simulation Setup · p. 4
- The onboard buffer starts at 400 Mb, is capped at 700 Mb, and receives data at ψ = 500 Mbps.The channel parameters are based on 28 GHz measurements, and maximum transmission power is 1 W.Source: A. Simulation Setup · p. 4
- NMPC uses horizons N = 30 in Scenario 1 and N = 40 in Scenario 2, with an obstacle safety radius of 0.35 m.The high-frequency CBF safety filter is implemented using CBFpy [15].Source: Figure 2 · p. 5
- Motion energy is evaluated with a discrete-time Pioneer 3DX motor-energy model, and the input and acceleration weights are W = diag(0.5, 0.1, 0.0) and Wa = diag(0.1, 0.1, 0.0).The same input and acceleration weights are used for the baselines; the sampling interval is 0.1 s.Source: A. Simulation Setup · pp. 4–5, Figure 2 · p. 5
B. Comparison with Baseline Methods
Across both scenarios, the proposed method lowers total energy relative to the motion and communication baselines. It does so by anticipating blockage, adapting transmission and routing to data rate, and limiting costly detours or NLOS transmission.
B. Comparison with Baseline Methods
- Total energy falls by 37.3% and 13.2% versus the motion and communication baselines in Scenario 1, and by 16.5% and 18.2% in Scenario 2, respectively.Communication energy is reduced by orders of magnitude when NLOS regions are unavoidable.Source: B. Comparison with Baseline Methods · p. 5
- In Scenario 1, the controller empties the buffer before NLOS regions and detours through an LOS region, unlike the motion baseline, which continues transmitting in NLOS.The motion baseline's transmission power exceeds the 1 W maximum allowable limit in NLOS regions.Source: B. Comparison with Baseline Methods · p. 5, Figure 3 · p. 6
- In Scenario 2, the method limits exposure to one NLOS region with a minor detour, while the baselines incur either costly NLOS transmission or larger motion detours.It avoids transmission in the NLOS region by preemptively emptying the buffer.Source: B. Comparison with Baseline Methods · p. 5
- As ψ rises from 100 to 1000 Mbps, the route shifts from tolerating NLOS regions to detouring around them to prevent buffer overflow.At ψ = 500 Mbps, the vehicle allows limited NLOS exposure while avoiding large trajectory deviations.Source: C. Impact of Data Accumulation Rate · p. 6, Figure 4 · p. 6
D. Computational Efficiency
The approach runs below the 100 ms sampling interval in both scenarios, while the CBF supplies safety guarantees that NMPC alone does not ensure. The conclusion reports energy savings of up to 37.3% in simulation.
D. Computational Efficiency
- Average solver times are 34 ms per step in Scenario 1 and 55 ms per step in Scenario 2, both below the 100 ms sampling interval.The paper describes the method as suitable for real-time applications.Source: D. Computational Efficiency · p. 6
E. Impact of Safety Filter (CBF)
- In Scenario 2 at ψ = 1000 Mbps, removing the CBF results in a collision with the rightmost obstacle.The CBF provides safety guarantees not ensured by NMPC alone, particularly under solver tolerances and real-time computation limits.Source: E. Impact of Safety Filter (CBF) · p. 6
VI. CONCLUSIONS
- In simulated dynamic mmWave environments, the NMPC framework with CBF reduces total energy by up to 37.3% versus baseline strategies.The conclusion also reports orders-of-magnitude lower communication energy when NLOS regions are unavoidable.Source: §VI · p. 6