10
GPUs Enable Deep Neuroevolution for Vision-Based Autonomous Driving Faustino Gomez, CEO NNAISENSE SA Lugano, Switzerland

GPUs Enable Deep Neuroevolution for Vision-Based …on-demand.gputechconf.com/gtc-eu/2017/presentation/2… ·  · 2017-10-27AUTOMATED DRIVING USING VISION Jan Koutnik, ... Evolving

Embed Size (px)

Citation preview

Page 1: GPUs Enable Deep Neuroevolution for Vision-Based …on-demand.gputechconf.com/gtc-eu/2017/presentation/2… ·  · 2017-10-27AUTOMATED DRIVING USING VISION Jan Koutnik, ... Evolving

GPUs Enable Deep Neuroevolution for Vision-Based Autonomous

Driving

Faustino Gomez, CEO NNAISENSE SA

Lugano, Switzerland

Page 2: GPUs Enable Deep Neuroevolution for Vision-Based …on-demand.gputechconf.com/gtc-eu/2017/presentation/2… ·  · 2017-10-27AUTOMATED DRIVING USING VISION Jan Koutnik, ... Evolving

SUPERVISED LEARNING

Learned Model

+Training set

prediction

target

errorinput

• Labeled training set

• Data is fixed and know a priori

• Learn from prediction error

• Regression / Classification

• Particularly amenable to GPUs

Page 3: GPUs Enable Deep Neuroevolution for Vision-Based …on-demand.gputechconf.com/gtc-eu/2017/presentation/2… ·  · 2017-10-27AUTOMATED DRIVING USING VISION Jan Koutnik, ... Evolving

REINFORCEMENT LEARNING

sense act

S1 S2 S3 Sn…

sense actsense act

a1 a2 a3

• Sequential decision task

• Data determined by learning agent

• No targets (i.e. teacher)

• Learn good control policy from sparse reward signal

• Partial obersvability

Environment

Learning Agent

reward

Page 4: GPUs Enable Deep Neuroevolution for Vision-Based …on-demand.gputechconf.com/gtc-eu/2017/presentation/2… ·  · 2017-10-27AUTOMATED DRIVING USING VISION Jan Koutnik, ... Evolving

NEUROEVOLUTION

Environment

fitness

evaluate

sensors action

fitness

Evolutionaryalgorithm

Page 5: GPUs Enable Deep Neuroevolution for Vision-Based …on-demand.gputechconf.com/gtc-eu/2017/presentation/2… ·  · 2017-10-27AUTOMATED DRIVING USING VISION Jan Koutnik, ... Evolving

Neuroevolution: Advantages

• No linearity assumptions

• Can cope with high-dimensional input/output

• Can use history of sensor readings of unknown depth (short-term memory)

• Can incorporate arbitrary constraints

• Behavior is learned not programmed

• Does not require knowledge of what constitutes optimal performance, i.e. reference signal

Page 6: GPUs Enable Deep Neuroevolution for Vision-Based …on-demand.gputechconf.com/gtc-eu/2017/presentation/2… ·  · 2017-10-27AUTOMATED DRIVING USING VISION Jan Koutnik, ... Evolving

SUCCESSFUL TEST CASE A U T O M AT E D D R I V I N G U S I N G V I S I O N

Jan Koutnik, Giuseppe Cuccu, Juergen Schmidhuber, and Faustino Gomez (2013). Evolving Large-Scale Neural Networks for Vision-Based Reinforcement Learning. In Proceedings of the Genetic and Evolutionary Computation Conference (GECCO, Amsterdam).

Million weight RNN learns to drive car using vision WITHOUT A TEACHER

Page 7: GPUs Enable Deep Neuroevolution for Vision-Based …on-demand.gputechconf.com/gtc-eu/2017/presentation/2… ·  · 2017-10-27AUTOMATED DRIVING USING VISION Jan Koutnik, ... Evolving

AUDI:Autonomous Parking

Objective: Learn to park car elegantly in more general conditions using only local sensors

• Reinforcement-learn continuous non-linear control • No global information • High-dimensional, noisy input • Closing reality gap (forward model) • Computationally intensive (physical model on GPU) • Timeline: build system for scratch in 6 months for NIPS

Challenges:

Page 8: GPUs Enable Deep Neuroevolution for Vision-Based …on-demand.gputechconf.com/gtc-eu/2017/presentation/2… ·  · 2017-10-27AUTOMATED DRIVING USING VISION Jan Koutnik, ... Evolving

AUDI PLAYCAR DEMMO

Page 9: GPUs Enable Deep Neuroevolution for Vision-Based …on-demand.gputechconf.com/gtc-eu/2017/presentation/2… ·  · 2017-10-27AUTOMATED DRIVING USING VISION Jan Koutnik, ... Evolving

Couldn’t do without GPU

• GPUs used to train all components

- CNN localizer

- Forward car model

- RNN controller

• Enable running 20K simulations parallel

• 100x speedup over multicore CPU cluster

Page 10: GPUs Enable Deep Neuroevolution for Vision-Based …on-demand.gputechconf.com/gtc-eu/2017/presentation/2… ·  · 2017-10-27AUTOMATED DRIVING USING VISION Jan Koutnik, ... Evolving

THANK YOU