Near-perfect point-goal navigation from 2.5 billion frames of experience

This could very well be a breakthrough by Facebook AI! However, 2.5 billion frames appears to be a bit high for this task!

"DD-PPO’s [decentralized distributed proximal policy optimization] success by creating systems that accomplish [indoor] point-goal navigation with only camera input — and no compass or GPS data"
Why no compass? The article contradicts itself about using or not using a compass!

"DD-PPO-trained agents reach their goal 99.9 percent of the time. Perhaps even more impressive, they do so with near-maximal efficiency, choosing a path that comes within 3 percent (on average) of matching the shortest possible route from the starting point to the goal. ... There is no scope for mistakes of any kind — no wrong turn at a crossroads, no backtracking from a dead end, no exploration or deviation of any kind from the most direct path."

Some challenges:
"Thus, existing distributed RL architectures do not scale and there is a need to develop a new distributed architecture."

"We propose a simple, synchronous, distributed RL [reenforcement learning] method that scales well. We call this method decentralized distributed proximal policy optimization, as it is decentralized (has no parameter server) and distributed (runs across many different machines), and we use it to scale proximal policy optimization, a previously developed technique ( Schulman et al., 2017 ). In DD-PPO, each worker alternates between collecting experience in a resource-intensive, GPU-accelerated simulated environment and then optimizing the model. This distribution is synchronous — there is an explicit communication stage in which workers synchronize their updates to the model."

"We used DD-PPO to train an agent for point-goal navigation for 2.5 billion steps (the equivalent of 80 years of human experience). This represented more than six months of GPU-time training, but we completed it in less than three days of wall-clock time with 64 GPUs."
This is still a massive effort!

Near-perfect point-goal navigation from 2.5 billion frames of experience: Facebook AI has effectively solved the task of point-goal navigation by AI agents in simulated environments, using only a camera, GPS, and compass data....