IEEE Robotics & Automation Magazine - June 2014 - 56

landmarks (Figure 1). It is assumed that the number of false
measurements generated by each vehicle is Poisson distributed
and possesses a uniform spatial distribution in the sensor FOV.
The probability of detecting a landmark present in the sensor
FOV is assumed to be constant and set to 0.99, and the probability of survival of a landmark from the current time step to
the next is also assumed to be constant and is set to 0.99. In the
RB-CMSLAM algorithm, the log-odds ratio is incremented in
steps of 0.01, and a threshold value of 0.03 is chosen for a feature to be confirmed as valid. These values are chosen to
achieve a comparable performance across varying clutter conditions and to cater for the miss detections. The number of
particles used is 100, and the particle resampling is performed
as the number of effective particles drop below 75.
The simulations are carried out under low (m (cr) = 1), mild
(r)
(m c = 5), and high (m (cr) = 10) clutter conditions, and the
performance of the RFS-GEN-CMSLAM algorithm is benchmarked against the RB-CMSLAM algorithm. It is apparent
that, although the RB-CMSLAM solution produces smaller

20

20

18

18

16

16

Positional RMSE (m)

Positional RMSE (m)

Simulation Results
The performance of the RFS CMSLAM algorithm based on
the general multisensor update method is evaluated using a
simulation, and the results are benchmarked against the
CMSLAM algorithm proposed by [9]. Herein, we refer to
Howard's solution as RB-CMSLAM and the RFS CMSLAM
solution based on the general multisensor update method as
RFS-GEN-CMSLAM. The RB-CMSLAM algorithm is chosen
as a better candidate for performance comparison since both
algorithms share the same standard particle filter-based vehicle
trajectories estimation method, while differentiating from the
map and measurement representation and the map propagation method. To improve the performance under measurement clutter, the RB-CMSLAM algorithm is combined with a
real-time implementation of the feature elimination method
based on the negative evidence strategy proposed by [2].
Two vehicles with identical control and sensor parameters
(given in Table 1) are driven on two different trajectories in a
simulation environment consisting of 41 randomly placed

14
12
10
8
6
4
2
0

14
12
10
8
6
4
2

0

0

500 1,000 1,500 2,000 2,500 3,000 3,500 4,000
Time Step (dt)

0

(a)

500 1,000 1,500 2,000 2,500 3,000 3,500 4,000
Time Step (dt)
(b)

60

60

50

50
Running Time (s)

Running Time (s)

Figure 2. The RMS vehicle position errors of 10 MC runs under low (in green), mild (in blue), and high (in red) clutter conditions. The
dotted graphs correspond to the first vehicle, while the nondotted graphs correspond to the second vehicle. (a) RB-CMSLAM.
(b) RFS-GEN-CMSLAM.

40
30
20
10
0

40
30
20
10

0

100 200 300 400 500 600 700 800 900 1,000

0

0

100 200 300 400 500 600 700 800 900 1,000

Time Index

Time Index

(a)

(b)

Figure 3. The average running time (in seconds) of each measurement update step, under low (in green), mild (in blue), and high
(in red) clutter conditions. (a) RB-CMSLAM. (b) RFS-GEN-CMSLAM.

56

*

IEEE ROBOTICS & AUTOMATION MAGAZINE

*

June 2014
Table of Contents for the Digital Edition of IEEE Robotics & Automation Magazine - June 2014
