Cognitive Intelligence, Deep Learning, Interviews

Changan Interview: Collaborative Perception and V2V enabled Platooning

Radovan Miucic
Senior Intelligent Vehicle Engineer, Changan US R&D Center, Inc.

Radovan Miucic
Senior Intelligent Vehicle Engineer, Changan US R&D Center, Inc.

Prior to the Auto.AI, we.CONECT spoke with Radovan Miucic, Senior Intelligent Vehicle Engineer at Changan US R&D Center, Inc.

we.CONECT: What are your main responsibilities in your current role?

Radovan Miucic: I am technical specialist for the Connected and Automated vehicle Research and Development at Changan US R&D Center, Inc. I lead a team of engineers responsible for researching and developing innovative cooperative applications for autonomous and advanced driver safety. Our work includes V2X applications, collaborative perception and platooning.

we.CONECT: What fascinates you most about autonomous driving?

Radovan Miucic: Autonomous driving is cutting edge technology involving many challenging and complex problems that are really fun to work on. It holds a promise of increased personal mobility and free time for other activities. Autonomous driving will not only be safer but also will contribute to the wellbeing of the travellers i.e. reducing stress during daily work commute.

we.CONECT: What are your predictions for autonomous driving?

Radovan Miucic: Traditional automotive OEMs will roll out automated features in the restricted access roads first. Robo-taxi companies will roll out the fully automated cars in more challenging environments. The two groups approach the problem from different angle. For robo-taxis, it may be acceptable to include very expensive sensors needed for perception and redundancy, however traditional OEMs will pay attention to cost conscious decision on the type of sensors going into the production.

we.CONECT: What role does AI and cognitive computing play in self-driving car technologies?

Radovan Miucic: AI plays very important role in the perception and path planning part of the autonomous development. At the moment, AI is not so much part of the actual controls of the vehicle. More and more data will be available and we will witness an expansion of the AI based solutions in every domain including automotive.

we.CONECT: There is a lot of confusion around autonomous driving. What are the different routes to level 5 and what are the biggest challenges to get there in the short term (in the next 5 years)?

Radovan Miucic: A human, as a driver, is very adaptable to sudden changes. Dealing with changes in weather, road surface and traffic conditions comes natural to us. A machine, even AI based, is good at well-defined tasks and perhaps less adaptable to sudden challenges. Level 5 in California on a sunny day with a well mapped road is one thing but a vehicle in rural Michigan during winter is another.

we.CONECT: How is your company developing deep learning capabilities? What are the challenges?

Radovan Miucic: We are collecting data and exploring different neural network designs for various vehicle functions. Some of the challenges we are facing are acceptance of the AI technology, technical difficulties, and its high development costs.

we.CONECT: Please explain in brief the key aspects of your session at the Auto.AI 2018.

Radovan Miucic: Firstly, as a brief introduction and food for thought, I would like to talk about current and ongoing research I am involved in Changan. These include the Collaborative Perception and V2V enabled Platooning.

Secondly, I would like to ask my peers questions and start discussion about:

1. The path to autonomous/automated transportation: Automotive OEMs/Tier 1 suppliers versus Software/Robotaxi companies

– Automotive OEMs/Teir1 suppliers tend to develop autonomous features gradually, introducing partial automation (L2 and L3). Software/Robotaxi companies are (skipping L2 and L3) starting their development from fully automated levels (L4 and L5). The two approaches have different requirements for sensing capabilities.
– Which approach is more viable?
– Will this mixture of autonomous levels be an issue?
– How will our manual driving be affected with various autonomous vehicles sharing the road?

2. Autonomous vehicle deployment:
Automotive OEMs/Tier 1 suppliers tend to limit the autonomous features to highways and restricted access roads. Software/Robotaxi companies are trying to solve autonomous driving for all environments including urban and suburban roads.

– What is likely to be near term deployment path?
– Dedicated lanes on highways?

3. Performance of the Artificial Intelligence can match or exceed the classical methods. AI is already part of the sensing systems. For example, performance of AI in classification of the objects form camera sensors typically outperforms any equations based methods. Similar findings could be said about vehicle controls.

– Would you trust AI based vehicle controller?
– What would it take for the automotive industry to accept AI based controls?
– How to certify AI based controls?
– Does functional safety (e.g. ISO 26262) still apply to AI based solutions? What about systems that support online learning?

we.CONECT: What expectations do you have towards the Auto.AI 2018?

Radovan Miucic: I expect to experience a positive and interactive environment with peers, automotive suppliers, and academics. I would like to network with similarly minded people, present some insides in Changan R&D work and listen to other people’s perspectives on the Connected and Autonomous Vehicle research and development approaches.

we.CONECT: Thank you for taking part in this interview.