a self-supervised terrain roughness estimator for off-road autonomous driving

A Self-Supervised Terrain Roughness Estimator for Off-Road Autonomous Driving

David Stavens and Sebastian Thrun

Stanford Artificial Intelligence Lab

David Stavens, Sebastian Thrun

Self-Supervised Learning

“Combines” strengths of multiple sensors.

Ultra-Precise, No Range Precise, Long Range


Overview

Introduction and Motivation Classifying Terrain Roughness Self-Supervised Learning Experimental Results


2005 DARPA Grand Challenge


Velocity Planning for DGC 2005

Mobile robotics traditionally focuses on steering.

But speed is also important.– Beyond stopping distance and lateral maneuverability.

For Grand Challenge 2005, our vehicle adapted its speed to terrain conditions, minimizing shock:– Increases electrical and mechanical reliability.– Mitigates pose error for laser projection.– Increases traction for improved maneuvers.– Seems to be correlated with slowing on “hard” terrain.


Velocity Planning for DGC 2005

Simple three state algorithm:– Drive at speed limit until shock threshold exceeded.– Slow to bring the vehicle within the shock threshold.

• Uses approx. linear relationship between shock and speed.• Which is also important for the new work we present.

– Accelerate back to the speed limit.

Discontinuous control problem.– Hard to solve with conventional control approaches.

We used supervised learning.


Experiments for DGC 05


This Talk: Next Logical Step

We expand our online approach to be proactive.– Our previous approach was entirely reactive.

Difficult to be that precise with laser scanners.– Hence problems of uncertainty and learning.

Accuracy required for roughness detection exceeds that required for obstacle avoidance.– 15cm vs. 2-4cm


Other Approaches to Velocity Control

Terramechanics: guidance through rough terrain.– Online assessment only at low speeds.– High speeds require a priori maps.

Our approach is both online and at high speeds.– Speeds up to 35 mph.


CMU’s Preplanning Trailer


Overview



Acquiring a 3D Point Cloud


Errors in Pose and Projection


Z Error vs. Time


More than t

“Spread” of plot implies more factors than t.

t is also related to:– Amount/rate of pitching.– Distance between the two scans.


Comparing Two Laser Points

pair =1| z |2 –

3| t |4 –

5| xy distance |6 –

7| dpitch1 |8 – 7| dpitch2 |8 –

9| droll1 |10 – 9| droll2 |10

Seven Features: z, t, xy distance, dpitches, drolls 10 Parameters: 1 2 … 10 (generated with self-supervised learning)


Combining Multiple Comparisons

n pairs in ascending order.– Use weighting because resolution of discontinuities is near

resolution of laser. There are not many witness pairs.

n

R = pair 11i

i = 0

This generates a score, R, for that patch of terrain.

But how do we assign target values to R?


Overview



Self-Supervised Learning

Actual shock when driving over terrain modifies belief about original laser scan.

Improves classifier for subsequent scans!


Caveat: Must Correct for Speed


Mapping from R to Shock

Learn a simple suspension model in parallel with the classifier:

Rcombined = Rleft 12 + Rright

12

Rleft and Rright is for the terrain under each wheel.


Overview



Summary

Road shock provides ground truth for previously perceived patches of road.

Perception model improves in real-time.

Future terrain assessment is more precise.

A faster route completion time is possible.– For the same amount of shock.

Works either “offline” or “as you drive.”– Offline results presented.


Questions?

a self-supervised terrain roughness estimator for off-road autonomous driving

Documents

sebastian thrunacquiring

sebastian thrunerrors

sebastian thruncomparing

sebastian thrunexperiments

sebastian thrunmore

timedavid stavens

projectiondavid stavens

cmdavid stavens