Background

Reliable detection and state estimation of road users (cars, bikes, pedestrians, …)  is critical to navigation in urban driving environments. In this project, our team will extend the autonomy pipeline for detection and explore multi-sensory methods for 3D object detection using sensing modalities such as vision and range sensors.

Project Objectives 

The objectives of this project are as follows:

1. Evaluate various state-of-the-art strategies for 3D object detection using radar and camera, etc.
2. Evaluate real-time capabilities of proposed method
3. Incorporate model into a full autonomy stack
4. Evaluate method on real-time and full scale system

Preferred Skills

ROS, Python, C++, Deep Learning/ML (3D object detection), 3D computer vision, experience with LIDAR

Project Timeline: at least 2 quarters

Background

Understanding the future states (i.e. 3-10 seconds) of road participants plays an important role in decision-making and navigation. In this project, several long term forecasting and intent recognition strategies will be considered and incorporated into our autonomy stack.

Project Objectives 

The objectives of this project are as follows:

1. Evaluate state of the art strategies for prediction
2. Evaluate real-time capabilities of proposed method
3. Incorporate model into a full autonomy stack
4. Evaluate method on real-time and full-scale system

Preferred Skills

ROS, Python, C++, ML (prediction), probabilistic state estimation and tracking

Project Timeline: at least 2 quarters

Background

Decision making entails defining a sequence of actions given spatiotemporal information about surrounding agents and obstacles provided by a perception stack. In this project, appropriate behaviors and decision-making strategies will be incorporated into our autonomy stack. Early testing and validation will be performed in simulation; deployment will be performed on full-scale vehicles in an urban setting.

Project Objectives 

The objectives of this project are as follows:

1. Implement a method for dynamic obstacle avoidance using learning based strategies
2. Benchmark regarding existing obstacle avoidance method
3. Evaluate real-time capabilities of proposed method
4. Incorporate model into a full autonomy stack
5. Evaluate methods for real-time and full-scale system

Preferred Skills

ROS, Python, C++, ML (imitation learning), graphical models (factor graphs).

Project Timeline: at least 2 quarters

Background

Scene understanding and state estimation of various agents depends on an unified representation of the scene. To facilitate this process, depth estimation and coordinate transformations from a sensor centric to an egocentric perspective are of relevance. In this project, several strategies will be considered for monocular depth estimation using image data and other sensing modalities to supervise the depth estimation process.

Project Objectives 

The objectives of this project are as follows:

1. Evaluate various state of the art strategies for monocular depth estimation and improve on existing techniques
2. Evaluate real-time capabilities and practicality of proposed method

Preferred Skills

Python, C++, Deep Learning/ML (computer vision), structure from motion, multi-view geometry

Project Timeline: 1-2 quarters