Skip to content

Commit fd1e523

Browse files
mrmonopoly-cybermrmonopoly-cyber
committed
docs(report): added physical experiments
1 parent 320cc2c commit fd1e523

File tree

6 files changed

+111
-6
lines changed

6 files changed

+111
-6
lines changed

‎doc/report/crazyflie_data_log/4_drones_all_together.txt‎

Lines changed: 3 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,3 @@
1+
6.40
2+
6.70
3+
6.40

‎doc/report/crazyflie_data_log/4_drones_reverse.txt‎

Lines changed: 3 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,3 @@
1+
38.58
2+
40.23
3+
39.56

‎doc/report/crazyflie_data_log/4_drones_sequentially.txt‎

Lines changed: 3 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,3 @@
1+
true
2+
true
3+
true

‎doc/report/crazyflie_data_log/distant_leader.txt‎

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1 @@
1+
10 m stable

‎doc/report/report.tex‎

Lines changed: 0 additions & 6 deletions
Original file line numberDiff line numberDiff line change
@@ -117,12 +117,6 @@
117117

118118
\input{sections/validation/validation}
119119

120-
\drawplot{4}
121-
\drawplot{8}
122-
\drawplot{12}
123-
\drawplot{16}
124-
\drawplot{20}
125-
126120
\input{bib}
127121

128122
\end{document}

‎doc/report/sections/validation/validation.tex‎

Lines changed: 101 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -60,3 +60,104 @@ \subsection{Limitations}
6060
Future work includes extending the simulation to dynamic scenarios and performing a more detailed
6161
quantitative analysis of message complexity.
6262

63+
\drawplot{4}
64+
\drawplot{8}
65+
\drawplot{12}
66+
\drawplot{16}
67+
\drawplot{20}
68+
69+
\subsection{Physical Deployment Validation}
70+
71+
To complement the simulation-based validation, the NearestAP protocol was also evaluated on a
72+
small-scale physical deployment consisting of four autonomous aerial nodes. The objective of these
73+
experiments was not to derive statistically rigorous performance metrics, but to verify that the
74+
qualitative convergence properties observed in simulation also manifest under real-world operating
75+
conditions.
76+
77+
The physical deployment introduces factors that are difficult to model accurately in simulation,
78+
including uncontrolled wireless interference, non-deterministic scheduling effects, clock drift,
79+
asymmetric connectivity, and environmental disturbances. Successful convergence under these
80+
conditions provides evidence that the observed protocol behavior is not an artifact of the
81+
simulation model.
82+
83+
\subsubsection{Testbed Description}
84+
85+
Each drone executed the same NearestAP implementation used in the simulated experiments, without any
86+
protocol-level modifications. Nodes communicated exclusively through broadcast wireless messages and
87+
operated without any form of global clock synchronization.
88+
89+
The deployment consisted of four drones operating concurrently. Convergence time was measured
90+
manually using wall-clock timing, with an estimated resolution on the order of one second. Each
91+
experiment was repeated approximately three times. No experimental run resulted in a failure to
92+
converge.
93+
94+
\subsubsection{Constant Potential Experiments}
95+
96+
In the first set of experiments, all nodes were assigned a fixed and identical potential. These tests
97+
were designed to validate correct leader convergence and recovery behavior under different startup
98+
and disturbance scenarios.
99+
100+
\paragraph{Simultaneous Startup}
101+
102+
All four drones were powered on simultaneously, representing a worst-case initial condition with
103+
maximal election contention. Convergence times of 6.40~s, 6.70~s, and 6.45~s were observed across three
104+
runs. In all cases, the protocol converged rapidly to a single stable leader.
105+
106+
\paragraph{Progressive Leader Removal}
107+
108+
Starting from a converged configuration, drones were powered off sequentially, beginning with the
109+
current leader. This experiment resulted in significantly longer convergence times of 38.58~s,
110+
40.23~s, and 39.56~s.
111+
112+
The increased convergence time is expected, as the node selected for shutdown at each step was the
113+
current leader. This configuration intentionally forces repeated leadership revocation and re-election,
114+
representing a highly adversarial scenario rather than a steady-state failure condition.
115+
116+
\paragraph{Sequential Startup}
117+
118+
Drones were powered on sequentially to evaluate whether late-joining nodes correctly recognize and
119+
defer to an already established leader. While precise convergence timings were not recorded for this
120+
scenario, all runs resulted in successful convergence without leadership instability.
121+
122+
\paragraph{Leader Isolation}
123+
124+
After convergence, the leader drone was physically displaced to a location more than approximately
125+
20~meters away from the other nodes, resulting in degraded and asymmetric connectivity. Despite this
126+
disturbance, the isolated leader remained stable for the full duration of a 10-minute observation
127+
period, with no leadership revocation or split-brain behavior observed.
128+
129+
\subsubsection{Battery-Dependent Potential Experiment}
130+
131+
To evaluate protocol behavior under heterogeneous and dynamic node conditions, an additional
132+
experiment was conducted in which node potential was influenced by battery state.
133+
134+
One drone was intentionally kept at a lower battery level than the others while remaining
135+
continuously powered via a USB connection to a host computer. This configuration ensured that the
136+
drone operated under constrained energy conditions while remaining active throughout the experiment.
137+
138+
Despite its reduced battery state, the low-energy drone eventually became leader. This confirms
139+
that leadership selection is governed by the protocol-defined potential mechanism rather than by
140+
startup order, hardware performance, or transient communication advantages.
141+
142+
\subsubsection{Observations}
143+
144+
Across all physical experiments, the protocol consistently converged to a single leader without
145+
manual coordination. No persistent split-brain conditions or permanent livelock scenarios were
146+
observed, even under repeated leader removal or physical isolation.
147+
148+
While convergence times were generally longer and more variable than those observed in simulation,
149+
the qualitative dynamics of candidate dominance, leadership stabilization, and recovery closely
150+
matched those observed in the simulated environment. In particular, the protocol demonstrated robust
151+
behavior under adversarial conditions specifically designed to disrupt stable leadership.
152+
153+
\subsubsection{Limitations}
154+
155+
The physical validation was conducted on a limited number of nodes and with a small number of
156+
repetitions. Convergence times were measured manually and are therefore subject to observer error and
157+
limited temporal resolution.
158+
159+
Furthermore, the experiments do not provide coverage of large-scale deployments or worst-case
160+
wireless interference patterns. As such, the physical validation should be interpreted as
161+
qualitative confirmation of protocol behavior rather than as a quantitative performance evaluation.
162+
163+

0 commit comments

Comments
 (0)