Isolation-aware Scheduling Framework for DNN-based End-to-End Autonomous Driving System on Tile-based Accelerators

2026-06-08

AI Summary

This work addresses the challenge of scheduling multiple heterogeneous deep neural networks (DNNs) in high-level autonomous driving systems under stringent end-to-end latency constraints, where resource contention and execution time variability pose significant obstacles to mass production. The paper introduces ADS-Tile, a novel framework that unifies tile-level hardware-native isolation with elastic parallelism into a single scheduling space, enabling spatiotemporally controllable isolation-sharing mechanisms. By integrating DAG-aware task modeling, probabilistic latency analysis, and elastic resource reservation, ADS-Tile jointly optimizes task co-location and parallelism allocation. Experimental results on an industry-academia collaborative benchmark demonstrate that, compared to work-conserving baselines, ADS-Tile reduces tile utilization by up to 32% in critical deadline scenarios and lowers computation waste caused by reallocation from 17%–44% to below 1.2%.

Abstract

Level-4+ autonomous driving systems (ADS) must run dozens of heterogeneous deep neural networks (DNNs) as end-to-end (E2E) pipelines under a strict latency constraint (<=100 ms), even as execution time varies by up to 3.3x. Cost rules out dedicating isolated hardware to each function in mass-produced ADS, so these DNNs must be densely colocated on a single chip, which introduces shared-resource contention. Tile-based accelerators expose two scheduling opportunities that conventional ADS schedulers do not exploit. First, they provide a tunable degree of parallelism (DoP): assigning more tiles raises DoP and can shorten DNN execution time. Second, they provide hardware-native isolation: tiles can be physically partitioned among co-located DNNs. But using this flexibility is expensive: changing a task's DoP triggers a stop-migrate-restart reallocation of its weights and intermediate features. At ADS task rates of 10-240 Hz, these stalls accumulate along E2E chains and threaten deadlines. Reservation-based schedulers fix DoP and leave this flexibility unused; work-conserving schedulers exploit it but assume reallocation is cheap and treat deadlines as independent. We present ADS-Tile that combines configurable isolation and elastic reservation into a spatio-temporal isolation-sharing space that bounds where and when reallocation occurs; a probabilistic latency model and a DAG-aware runtime scheduler then use this space to decide task colocation and DoP under shared E2E deadlines. On an industry- and academia- derived ADS benchmark, ADS-Tile uses up to 32% fewer tiles than the work-conserving baseline in deadline-critical settings and cuts reallocation-induced wasted processing capacity from 17%-44% to below 1.2%. Controlled spatio-temporal sharing improves resource efficiency and latency predictability for tile-based ADS.

Problem

• autonomous driving
• DNN scheduling
• tile-based accelerators
• resource contention
• reallocation overhead

Innovation

• tile-based accelerators
• isolation-aware scheduling
• elastic reservation
• spatio-temporal sharing
• DNN co-location