Cadence Collaboration with Kudan and Visionary.ai Enables Rapid Deployment of VSLAM and AI ISP-Based Solutions

Are you confused while navigating new environments, especially in less optimum light conditions? 

I am, and the problem worsens when lighting conditions are lousy and GPS connectivity is poor! Then, I try to locate my destination by looking at some landmarks.  

Likewise, robots and self-driving vehicles use simultaneous localization and mapping (SLAM) and digital imaging to navigate unknown environments, even in the worst lighting conditions. SLAM and digital imaging work as an embedded vision for robots and self-driving vehicles as they make maps (especially while working indoors) of unknown environments while navigating through them.   Such advancements incorporate different kinds of sensors and computer vision-based algorithms for SLAM, 3D object detection, tracking, and trajectory estimation. These computational imaging algorithms are very complex, involve extensive computational resources, and may lead to increased latency and power requirements. Digital Imaging is another major trend plagued by challenges such as low light, high/wide dynamic range (HDR/WDR) environments, and fast-moving objects. These are the most significant barriers to clear and crisp imaging. To best address these challenges, Cadence has signed up with Kudan and Visionary.ai for its Tensilica software partner ecosystem, bringing SLAM and AI image signal processor (ISP) to its processor cores. This partnership helps achieve the best performance in various segments such as advanced automotive, mobile, consumer and IoT, and drones. The proof is in the pudding; Cadence strengthens Tensilica Vision and AI software partner ecosystem for advanced automotive, mobile, consumer, and IoT Applications. While Tensilica Vision Q7 DSP helped achieve a nearly 15% speedup of Kudan's proprietary SLAM implementation pipeline compared to CPU-based implementation, Tensilica NNA110 accelerator helps customers implement a camera pipeline with a resolution of more than full HD at over 30fps. 

How Do Robots/Self-Driving Vehicles Navigate? And Why Do We Need AI-Based Image Signal Processors (ISP)? 

Empowering self-driving vehicles and robots with the ability to make this world a safer place is the talk of the hour. It involves navigating unknown environments, producing video even in the worst lighting conditions, and processing at the edge. SLAM empowers robots and self-driving vehicles with an embedded vision that helps construct or update an unknown environment map for inside-out tracking with embedded vision. It is commonly used in the robotics, drone, mobile, AR/VR and automotive markets. SLAM involves computationally heavy and a variety of linear algebra and matrix operations that require more time and processing under the hood, and it comprises feature detection, descriptor matching perspective transformation various filters. 

GPUs are mostly floating-point machines, while a lot of these operations can now be done in fixed points, where DSP-type machines can benefit. So, you can get more parallel operations in the same register size with an integer. The usage of specialized processors such as GPU is limited due to performance capabilities and limitations such as high-power dissipation. The ISP functions are provided as hardware blocks in most systems with image/video functionality that offers high-performance throughput but suffers from issues like lack of flexibility or adaptability to current conditions. Also, hardware solutions with algorithms result in the loss of data captured by sensors. So, with AI now taking reign, image signal processors (ISP) can be found in a wide number of applications from smartphones to automotive and beyond.

How Cadence Tensilica Ecosystem Helps Automotive, Roots, IoT, and Mobile

A robust ecosystem (hardware and software) is essential to address the challenges mentioned above. Tensilica Vision and software ecosystem include more than 50 partners developing solutions for automotive, smartphone apps, IoT, software services and other segments. Cadence Tensilica IP-based devices help to run cutting-edge SLAM and AI ISP solutions efficiently and offer the best power-performance envelope. The ongoing innovations in Tensilica IP and architecture are critical for smartphone manufacturers and providers of IoT systems and next-generation connected vehicles. Cadence partnership with Industry leaders such as Kudan and Visionary.ai helps customers improve performance. Various benefits of these collaborations are as below:

1. Tensilica Vision Q7 DSP offers a 10X performance improvement and 15% speedup compared to CPU-based implementations of Kudan'c proprietary SLAM implementation.

"We're very excited about our partnership with Cadence and the opportunity to work with the Tensilica platform to accelerate Kudan's SLAM pipeline. Cadence's Tensilica Vision DSPs provide specialized instructions that optimize various stages of the SLAM algorithm, delivering significant gains with power savings to the end customer. We look forward to improving accessibility and adoption of our SLAM solution together."

-Juan Wee, CEO at Kudan USA

2. Visionary.ai's novel approach leverages AI to replace traditional hardwired ISP functions and enables real-time, high-quality video production, even in the most challenging lighting conditions.
3. Customers could implement a camera pipeline with a much higher resolution than full HD while working with over 30fps using Visionary.ai's efficient AI ISP over Tensilica NNA110.

"At Visionary.ai, we have developed a method of using AI to dramatically improve image quality in real-time, particularly in the most challenging lighting conditions. For this technology to reach its true potential, there is a need for fast and efficient neural network computations. Joining Cadence's Tensilica ecosystem will help ensure that our customers have a very competitive solution that runs on some of the most efficient vision and AI platforms out of the box."

 Oren Debbi, CEO at Visionary.ai

Key Features Leading to the Adoption of the Tensilica Q7 DSP

Tensilica Vision Q7 DSP, The Cadence Tensilica Q7 DSP, is designed to meet the needs of applications using SLAM, enabling high-performance SLAM on the edge and in other devices.

With optimized instructions for faster performance on matrix operations, feature extraction, and the capability to provide a perfect balance of high performance and low power (essential to SLAM applications at the edge) along with convolutions, Cadence Tensilica Q7 DSP is poised to give the best performance. Cadence's Tensilica Vision Q7 DSP (digital signal processor), designed to enhance computer vision and AI applications, optimizes Kudan's SLAM pipeline— providing customers with a versatile, high-performance computing platform with SLAM capabilities for applications such as robotics. Click here for more details about Tensilica Vision Q7 DSP.

Key Features Leading to the Adoption of the Tensilica Neural Network Accelerator (NNA) 110

Cadence partners with Visionary.ai to achieve cutting-edge imaging technology. Tensilica processors are used for many AI applications due to performance, power, and what types of data are required (floating point, 8-bit, and so on). The NNA products include random sparse compute to improve performance, run-time tensor compression to decrease memory bandwidth, pruning, and clustering to reduce model size.

The Cadence Tensilica NNA 110 accelerator incorporates a custom hardware accelerator engine (NNE) coupled with a Tensilica Vision P6 or P1 DSP. The specialized compute block inside the NNA 110 hardware leverages features like random sparsity and tensor compression/decompression to provide an overall best-in-class embedded AI accelerator solution. Click here to read more about Tensilica NNA110.

 Learn More

• Cadence Strengthens Tensilica Vision and AI Software Partner Ecosystem for Advanced Automotive, Mobile, Consumer, and IoT Applications
• Kudan Partners with Cadence to Enhance Visual SLAM Performance for Robotics Applications
• Vision Q7 DSP
• AI Max – NNA 110 Single Core