Germany bans Tesla from claiming its Autopilot software is potentially autonomous
Also: Create the killer AI Doom player on a workstation
In brief A judge in Munich has ruled in favor of banning Tesla Germany from repeating misleading descriptions of its Autopilot software in adverts.
Phrases like “full potential for autonomous driving” and "Autopilot inclusive" should be no longer allowed, Wettbewerbszentrale, an independent German regulatory institution, argued. It told the court the company's adverts could fool people into believing that Tesla's self-driving cars were fully autonomous and could drive without any human assistance. The technology is not capable of doing that.
Also, self-driving cars aren't allowed on Germany's road, yet Tesla's ads suggested otherwise.
Tesla describes its Autopilot software as requiring a “fully attentive driver, who has their hands on the wheel and is prepared to take over at any moment.” The automaker can appeal the decision if it wants to.
New Amazon self-driving toy car
If you want to carry out practical experiments on autonomous vehicles in a much cheaper and safer way, then you can now consider Amazon’s AWS DeepRacer Evo cars.
The minuscule robot cars come with stereo cameras and a LIDAR sensor, an upgrade from its original AWS Deepracer model. The gizmo also comes equipped with an Intel Atom CPU Processor, 4GM RAM and 32GB of storage to run machine learning software for self-driving.
Tinkerers can train and test reinforcement learning algorithms in simulation using the AWS DeepRacer application. Be aware, you will need to purchase cloud credits via Amazon Sagemaker to train the software. The cars can also race against one another virtually.
The AWS DeepRacer Evo is currently discounted at $399; it’s $598 at full price. If you have the older version, you can upgrade it using the AWS DeepRacer sensor kit that’s being sold right now at $149. Its original price is $249.
Intel and the University of Southern California claim to have invented a reinforcement learning (RL) algorithm that can run on a high-end workstation as opposed to on a cluster of machines.
In a paper presented at the 2020 International Conference on Machine Learning (ICML) this week, the team trained an AI be to be absolute killer at Doom, the 1990s first-person shooter that ignited the games industry.
“Inventing ways to do deep RL on commodity hardware is a fantastic research goal,” Peter Stone, a professor at the University of Texas at Austin told IEEE. “Any progress towards democratizing RL and reducing the energy needs for doing research is a step in the right direction.” ®