notesum.ai

Published at December 10

POMDP-Based Trajectory Planning for On-Ramp Highway Merging

cs.RO

Released Date: December 10, 2024

Authors: Adam Kollarčík¹, Zdeněk Hanzálek¹

Aff.: ¹Czech Technical University in Prague

Arxiv: http://arxiv.org/pdf/2412.07567v1

Refer to caption

Description

Value

A

action set

\left\{-1,-0.5,0,0.5,1\right\}

\mathrm{m}\text{\,}{\mathrm{s}}^{-2}

\times

\left\{-2,-1,-0.5,0,0.5,1,2\right\}

\deg

\Delta t

time step length

1

\sec

c

UCB parameter

200000

N

ABT depth

10

n_{\mathrm{ep}}

ABT number of episodes

10000

n_{\mathrm{par}}

ABT number of particles

1000

\gamma

discount factor

1

\bm{Q}

dynamics covariance

\bm{0}

\bm{\Sigma}

IDM covariance

\mathrm{diag}(9,0.04)

v_{\mathrm{des}}

desired velocity

27.8

\mathrm{m}\text{\,}{\mathrm{s}}^{-1}

a_{\mathrm{max}}

IDM maximal acceleration

1

\mathrm{m}\text{\,}{\mathrm{s}}^{-2}

a_{\mathrm{min}}

IDM minimal acceleration

-1

\mathrm{m}\text{\,}{\mathrm{s}}^{-2}

\tau

IDM time headway

1.5

\mathrm{s}

\delta

IDM acceleration exponent

4

d_{\mathrm{min}}

IDM minimal distance

1

\mathrm{m}

\bm{R}

observation covariance

\mathrm{diag}(0.01,0.01,0)

R_{\mathrm{center}}

centerline reward coef.

500

R_{\mathrm{acc}}

acceleration reward coef.

100

R_{\mathrm{steer}}

steering reward coef.

10\cdot(180/\pi)^{2}

R_{\mathrm{vel}}

velocity reward coef.

100

R_{\mathrm{crash}}

crash reward coef.

1000000

R_{\mathrm{cst}}

wrong lane constant reward coef.

1000

R_{\mathrm{end}}

wrong lane end reward coef.

10000

R_{\mathrm{dist}}

wrong lane distance reward coef.

500

R_{\mathrm{h}}

heuristics coef.

100