UC Berkeley – Which Mutual Information Representation Learning Objectives are Sufficient for Control?

Processing raw sensory inputs is crucial for applying deep RL algorithms to real-world problems. For example, autonomous vehicles must make decisions about how to drive safely given information flowing from cameras, radar, and microphones about the conditions of the road, traffic signals, and other cars and pedestrians. However, direct “end-to-end” RL that maps sensor data…

UC Berkeley – Bridge Data: Boosting Generalization of Robotic Skills with Cross-Domain Datasets

Fig. 1: The BRIDGE dataset contains 7200 demonstrations of kitchen-themed manipulation tasks across 71 tasks in 10 domains. Note that any GIF compression artifacts in this animation are not present in the dataset itself. When we apply robot learning methods to real-world systems, we must usually collect new datasets for every task, every robot, and…

UC Berkeley – Why Generalization in RL is Difficult: Epistemic POMDPs and Implicit Partial Observability

Many experimental works have observed that generalization in deep RL appears to be difficult: although RL agents can learn to perform very complex tasks, they don’t seem to generalize over diverse task distributions as well as the excellent generalization of supervised deep nets might lead us to expect. In this blog post, we will aim…

UC Berkeley – RECON: Learning to Explore the Real World with a Ground Robot

An example of our method deployed on a Clearpath Jackal ground robot (left) exploring a suburban environment to find a visual target (inset). (Right) Egocentric observations of the robot. Imagine you’re in an unfamiliar neighborhood with no house numbers and I give you a photo that I took a few days ago of my house,…

UC Berkeley – Making RL Tractable by Learning More Informative Reward Functions: Example-Based Control, Meta-Learning, and Normalized Maximum Likelihood

Diagram of MURAL, our method for learning uncertainty-aware rewards for RL. After the user provides a few examples of desired outcomes, MURAL automatically infers a reward function that takes into account these examples and the agent’s uncertainty for each state. Although reinforcement learning has shown success in domains such as robotics, chip placement and playing…

AI Trends – How Accountability Practices Are Pursued by AI Engineers in the Federal Government  

By John P. Desmond, AI Trends Editor    Two experiences of how AI developers within the federal government are pursuing AI accountability practices were outlined at the AI World Government event held virtually and in-person this week in Alexandria, Va.  Taka Ariga, chief data scientist and director, US Government Accountability Office Taka Ariga, chief data scientist and director at the US Government Accountability…

AI Trends – Startup: AssemblyAI Represents New Generation Speech Recognition 

By AI Trends Staff   Advances in the AI behind speech recognition are driving growth in the market, attracting venture capital and funding startups, posing challenges to established players.   The growing acceptance and use of speech recognition devices are driving the market, which according to an estimate by Meticulous Research is expected to reach $26.8 billion…

AI Trends – Pursuit of Autonomous Cars May Pose Risk of AI Tapping Forbidden Knowledge

By Lance Eliot, the AI Trends Insider     Are there things that we must not know?    This is an age-old question. Some assert that there is the potential for knowledge that ought to not be known. In other words, there are ideas, concepts, or mental formulations that should we become aware of that knowledge it could be…

AI Trends – Digital Natives Seen Having Advantages as Part of Government AI Engineering Teams 

By John P. Desmond, AI Trends Editor   AI is more accessible to young people in the workforce who grew up as ‘digital natives’ with Alexa and self-driving cars as part of the landscape, giving them expectations grounded in their experience of what is possible.   That idea set the foundation for a panel discussion at AI World…

AI Trends – Getting Government AI Engineers to Tune into AI Ethics Seen as Challenge 

By John P. Desmond, AI Trends Editor   Engineers tend to see things in unambiguous terms, which some may call Black and White terms, such as a choice between right or wrong and good and bad. The consideration of ethics in AI is highly nuanced, with vast gray areas, making it  challenging for AI software engineers to…