The Architectural Implications of Autonomous Driving

LinShih-Chieh,; ZhangYunqi,; HsuChang-Hong,; SkachMatt,; HaqueMd, E; TangLingjia,; MarsJason,

doi:10.1145/3296957.3173191

Cited by 62 publications

(12 citation statements)

References 41 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…From all AD software functionalities, our results on several widely-used state-of-the-art AD systems confirm the results of previous studies [15] showing that object detection is one of the most compute and energy-intensive modules. In this work, we focus on the camera-based object detection module of industrial autonomous driving frameworks such as Apollo [5] and Autoware [1], which operates on multiple cameras at a high frame rate (i.e.…”

Section: Introductionsupporting

confidence: 86%

“…Naturally, these hardware devices consume significant amounts of energy, which recent studies show can reduce the driving range (i.e. autonomy of cars) more than 10% [15]. This calls for hardware and software solutions to reduce the performance and energy requirements of AD software, without reducing the accuracy of the AD system due to its high criticality.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

IntPred

Tabani

Fusi

Kosmidis

et al. 2020

Proceedings of the 35th Annual ACM Symposium on Applied Computing

View full text Add to dashboard Cite

Deep Neural-Network (DNN) based Object Detection is one of the most important and time-consuming stages of Autonomous Driving software in cars. In non-critical domains, the performance and energy requirements of object detection can be reduced at the cost of accuracy in the detected objects. This is not the case in a critical domain like automotive, for which a delicate balance between performance/energy overheads and accuracy of object detection must be found. We propose IntPred to achieve such a balance by leveraging on the fact that, with high frame rates, objects do not move significantly across frames. IntPred tailors object interpolation for the case of object detection in autonomous driving frameworks, in line with approaches devised for other domains, thus heavily reducing the performance requirements of full-fledged DNN-based object prediction. IntPred results in comparable accuracy to the original object detection, while saving more than 70% of the computations. The latter allows using lower-performance and cheaper platforms resulting in saving energy and reducing heat dissipation: for instance, in an NVIDIA Jetson TX2 platform, specific for autonomous driving systems, our technique increases the frame processing rate by 4.6x. IntPred also allows consolidating additional applications onto the same platform.

show abstract

Section: Introductionsupporting

confidence: 86%

Section: Introductionmentioning

confidence: 99%

IntPred

Tabani

Fusi

Kosmidis

et al. 2020

Proceedings of the 35th Annual ACM Symposium on Applied Computing

View full text Add to dashboard Cite

show abstract

“…Smartphone companies are incorporating Artificial Intelligence (AI) chips in their design for ondevice inference to improve user experience and tighten data security, and the autonomous vehicle industry is turning to application-specific integrated circuits (ASICs) to keep the latency low. While the typical acceptable latency for real-time inference in applications like those above is O(1) ms [1,2], other applications may require sub-microsecond inference. For instance, high-frequency trading machine learning (ML) algorithms are running on field-programmable gate arrays (FPGAs) to make decisions within nanoseconds [3].…”

Section: Introductionmentioning

confidence: 99%

“…• We have implemented a range of quantization methods in a common library, which provide a broad base from which optimal quantizations can easily be sampled; • We introduce a novel method for finding the optimal heterogeneous quantization for a given model, resulting in minimum area or minimum power DNNs while maintaining high accuracy; • We have made these methods available online in easy-to-use libraries, called QKeras and Au-toQKeras 1 , where simple drop-in replacement of Keras [32] layers makes it straightforward for users to transform Keras models to their equivalent deep heterogeneously quantized versions, which are trained quantization aware. Using AutoQKeras, a user can trade-off accuracy by model size reduction (e.g.…”

Section: Introductionmentioning

confidence: 99%

Automatic heterogeneous quantization of deep neural networks for low-latency inference on the edge for particle detectors

et al. 2021

View full text Add to dashboard Cite

Although the quest for more accurate solutions is pushing deep learning research towards larger and more complex algorithms, edge devices demand efficient inference and therefore reduction in model size, latency and energy consumption. One technique to limit model size is quantization, which implies using fewer bits to represent weights and biases. Such an approach usually results in a decline in performance. Here, we introduce a method for designing optimally heterogeneously quantized versions of deep neural network models for minimum-energy, high-accuracy, nanosecond inference and fully automated deployment on chip. With a per-layer, per-parameter type automatic quantization procedure, sampling from a wide range of quantizers, model energy consumption and size are minimized while high accuracy is maintained. This is crucial for the event selection procedure in proton-proton collisions at the CERN Large Hadron Collider, where resources are strictly limited and a latency of O(1) µs is required. Nanosecond inference and a resource consumption reduced by a factor of 50 when implemented on field-programmable gate array hardware are achieved. FIG.I. An ultra-compressed deep neural network for particle identification on a Xilinx FPGA.

show abstract

“…The obtained results also indicate hp-DCFNoC is very well suited to applications with high-bandwidth requirements like the ones found in autonomous driving systems [12]. Note that performance guarantees of hp-DCFNoC are identical to the ones provided by DCFNoC while the performance that can be guaranteed in a 4×4 mesh wormhole NoC is much lower [4], [11].…”

Section: Analyzing the Impact On Applications Behaviourmentioning

confidence: 51%

HP-DCFNoC: High Performance Distributed Dynamic TDM Scheduler Based on DCFNoC Theory

et al. 2020

View full text Add to dashboard Cite

The need for increasing the performance of critical real-time embedded systems pushes the industry to adopt complex multi-core processor designs with embedded networks-on-chip. In this paper we present hp-DCFNoC, a distributed dynamic scheduler design that by relying on the key properties of a delayed conflict-free NoC (DCFNoC) is able to achieve peak performance numbers very close to a wormhole-based NoC design without compromising its real-time guarantees. In particular, our results show that the proposed scheduler achieves an overall throughput improvement of 6.9× and 14.4× over a baseline DCFNoC for 16 and 64-node meshes, respectively. When compared against a standard wormhole router 95% of its network throughput is preserved while strict timing predictability as property is kept. This achievement opens the door to new high performance time predictable NoC designs.

show abstract

The Architectural Implications of Autonomous Driving

Cited by 62 publications

References 41 publications

IntPred

IntPred

Automatic heterogeneous quantization of deep neural networks for low-latency inference on the edge for particle detectors

HP-DCFNoC: High Performance Distributed Dynamic TDM Scheduler Based on DCFNoC Theory

Contact Info

Product

Resources

About