# EcoFusion: Energy-Aware Adaptive Sensor Fusion for Efficient Autonomous Vehicle Perception

**Authors:** Arnav Vaibhav Malawade, Trier Mortlock, Mohammad Abdullah Al Faruque  
**arXiv:** [2202.11330](https://arxiv.org/abs/2202.11330)  
**Date:** 2022-02-23  
**Summary page:** https://www.summarizepaper.com/en/paper/2202.11330v1/

## Key points

- The paper proposes EcoFusion, an energy-aware sensor fusion approach for autonomous vehicles.
- EcoFusion leverages contextual information to adapt the fusion method and reduce energy consumption without compromising perception performance.
- The proposed approach outperforms existing fusion methods by up to 9.5% in object detection.
- EcoFusion consumes approximately 60% less energy and achieves a 58% lower latency on the Nvidia Drive PX2 hardware platform.
- The authors propose strategies for context identification and implement joint optimization between energy efficiency and performance.
- Scenario-specific results are provided to demonstrate the effectiveness of the approach.
- The research contributes to advancing autonomous vehicle perception by addressing energy consumption challenges without compromising performance.

## Summary

The paper titled "EcoFusion: Energy-Aware Adaptive Sensor Fusion for Efficient Autonomous Vehicle Perception" proposes a novel approach to sensor fusion in autonomous vehicles. The authors highlight that while autonomous vehicles rely on multiple sensors, large deep-learning models, and powerful hardware platforms to perceive the environment and navigate safely, some sensing modalities can negatively impact perception and increase energy consumption. To address this issue, the authors introduce EcoFusion, an energy-aware sensor fusion approach that leverages contextual information to adapt the fusion method and reduce energy consumption without compromising perception performance. The proposed approach outperforms existing fusion methods by up to 9.5% in object detection while consuming approximately 60% less energy and achieving a 58% lower latency on the industry-standard Nvidia Drive PX2 hardware platform. In addition to presenting EcoFusion, the authors also propose several strategies for context identification and implement a joint optimization between energy efficiency and performance. They provide scenario-specific results to demonstrate the effectiveness of their approach. The research contributes to advancing the field of autonomous vehicle perception by addressing the challenges of energy consumption without compromising performance. The paper is authored by Arnav Vaibhav Malawade, Trier Mortlock, and Mohammad Abdullah Al Faruque and has been accepted for publication in the 59th ACM/IEEE Design Automation Conference (DAC 2022).