This case study emphasizes  MulticoreWare’s role in Perception algorithm development with Automotive Radars for ADAS applications for one of our clients.

The client

The customer is a global technology company that develops sensors, sensor-based solutions for Automotive, Aerospace, Industrial, Off-road vehicles. The customer was developing autonomous navigation solutions for Off-road vehicles including its software processing pipeline.

Objective

To design an algorithm that uses Radar and IMU measurements to create a static map of the environment and localize the Ego-vehicle. For this, the static and moving objects in the environment needs to be identified. The static objects need to be represented using 3-dimensional bounding boxes. The localization process must estimate the position (in x,y,z coordinates) and orientation (roll, pitch, yaw) of the Ego-vehicle.

Constraints

Following were the constraints:

• The algorithm had to be tailored for Radar and IMU measurements obtained in an off-road scenario
• The Radar and IMU measurements were assumed not to be time-synchronized
• The computational complexity of the algorithm must be minimal

Sensors

4D Imaging Radar and IMU

Technology

Simultaneous Localization & Mapping

Scenario

Large-Vehicle, Off-Road Autonomous Machines

Approach

Multicoreware oversaw the R&D from ground up, and we evaluated three approaches to solve the customer’s problem. The methods being investigated were:

1. SLAM algorithm based on Particle filter and Kalman filter
2. Scan matching algorithm based on Iterative Closest Point Matching
3. Radar-Gyro fusion algorithm

Radar-Gyro fusion algorithm

• The preferred approach by the customer was implemented
• The approach had the best benefit in terms of accuracy vs latency trade-offs
• RADAR –For estimating the translational movement of Ego-vehicle and static mapping
• GYRO – For estimating the rotational movement of Ego-vehicle and its orientation