uSLAM Applications — Drones: AM970’s In-House Low-Power VPU Improves Obstacle Avoidance and Localization

uSLAM Applications — Drones:

AM970’s in-house high-performance, low-power VPU delivers more precise obstacle avoidance and reliable localization!

Real-time obstacle avoidance and environmental localization are among the hardest challenges in autonomous drone flight. Conventional solutions depend on power-hungry CPUs or costly sensors. AMicro’s machine-vision platform, uSLAM (ultra fusion SLAM), includes AM970, built around a dedicated vision processing unit (VPU) to deliver high-performance, low-power and low-cost intelligent perception.

0 1

AM970: a dedicated vision technology platform for intelligent drones

AM970

A dedicated vision technology platform for intelligent drones

Obstacle avoidance and localization are two core challenges when drones fly in complex environments:

How can obstacles be detected in real time during high-speed flight?

How can localization remain accurate with weak GPS signals?

What if a conventional CPU provides insufficient computing power and consumes too much energy?

AM970 addresses these challenges with a VPU optimized for machine vision, 720p@120fps depth output, SLAM acceleration and intelligent multi-sensor fusion, enabling clearer perception, faster processing and more stable flight.

0 2

AM970 Core Technologies:

Four advances in drone vision

1️⃣ Depth Engine: redefining obstacle avoidance and localization accuracy

Industry-leading 720p@120fps depth output enables real-time obstacle detection and precise hovering even during high-speed flight. Compared with conventional algorithms, point clouds have clearer edges, less noise and over 40% fewer holes. Performance advantages include:

Industry-leading 720p@120fps depth output enables real-time obstacle detection and precise hovering even during high-speed flight.

Compared with conventional algorithms, point clouds have clearer edges, less noise and over 40% fewer holes.

Key performance advantages:

120fps processing, compared with 8fps for the CPU solution

Almost no CPU resource consumption, allowing better resource allocation

✅ Comparison data:

2️⃣ Dedicated SLAM IP: reducing CPU use and improving localization stability

Designed for weak-GPS environments, customized SLAM IP substantially reduces CPU load. Measurements show nearly 4 times faster feature extraction (1.51ms vs 5.49ms), CPU utilization at just 18% of the conventional solution (0.14% vs 0.79%) and a higher RANSAC pass rate (93.06% vs 90.46%).

Designed for weak-GPS environments, customized SLAM IP substantially reduces CPU load.

Measurements show nearly 4 times faster feature extraction (1.51ms vs 5.49ms), CPU utilization at just 18% of the conventional solution (0.14% vs 0.79%) and a higher RANSAC pass rate (93.06% vs 90.46%).

Measured performance:

Nearly 4 times faster feature extraction (1.51ms vs 5.49ms)

CPU utilization at just 18% of the conventional solution (0.14% vs 0.79%)

Higher RANSAC pass rate (93.06% vs 90.46%)

✅ Additional benefit: a built-in precision clock synchronization unit aligns IMU and camera data, avoiding dropped frames under high system load.

3️⃣ Dedicated ISP: flexible tuning for different scenarios

Intelligent image processing provides customized imaging for conditions such as night flight and strong backlighting. ISP tuning tools and professional technical support simplify development.

Intelligent image processing provides customized imaging for conditions such as night flight and strong backlighting.

Developer-friendly ISP tuning tools and professional technical support simplify development.

4️⃣ Smart Fusion: balancing perception performance, cost and power

Multi-source fusion intelligently combines third-party dense point clouds with conventional stereo point clouds, controlling cost and power while preserving accuracy.

Multi-source fusion intelligently combines third-party dense point clouds with conventional stereo point clouds.

Cost and power are controlled while preserving accuracy.

0 3

AM970 Hardware:

Built for AI perception in drones

✔ Four physical MIPI interfaces, each supporting Virtual Channels, can connect up to 10 image sensors through expansion. ✔ 5T@INT8 computing supports more than 120 operators, weight sparsity and custom operators. ✔ High performance, low power consumption, safety and stability, with volume shipments already underway.

0 4

The Future Is Here:

AM970 redefines intelligent flight

✦ For brands and solution providers: a cost-effective perception solution lowers barriers to advanced vision, accelerates product upgrades and supports market competitiveness. ✦ For developers and enthusiasts: 5T computing, a flexible ISP and multi-sensor support create an adaptable environment for rapid AI deployment. ✦ For users: more stable flight, efficient obstacle avoidance and localization, and smarter interaction deliver an improved experience.

AM970 is more than a chip: its innovative VPU architecture provides a complete drone perception solution combining performance and cost advantages, advancing autonomous flight in complex environments.