AI Explorer Session

Iteration 1

Strategy focuses on traffic light timing variations to reduce delays by allocating more green time to higher-flow east-west direction, while testing safety impacts through yellow and all-red time adjustments. Experiments cover baseline, balanced changes, directional preferences, and extremes to explore the parameter space systematically.

Analysis (Iteration 1)

The results show that balanced green times (25s each) yield the lowest average waiting time of 103.53, while highly unbalanced configurations (e.g., 20/45 or 60/10) cause significantly higher delays up to 552s. All configurations maintain perfect safety with zero collisions and emergency braking, indicating that shorter cycle times and balanced allocation are key for efficiency without compromising safety.

Key Insights:

• Balanced green times (ew_green_time ≈ ns_green_time) consistently perform better than unbalanced ones, with the 25/25 configuration achieving the best results.
• Shorter cycle times (around 60-75s) correlate with lower waiting times, while longer cycles (90s) increase delays significantly.
• Yellow and all_red times have minimal impact on waiting time, with variations from 2-5s and 1-4s respectively showing no clear trend.
• Safety metrics remain at zero across all configurations, suggesting that reducing waiting times doesn't compromise collision or emergency braking risks.

Recommendations:

• Explore more balanced green time configurations around 25-30s each, potentially testing slight asymmetries based on traffic demand.
• Experiment with shorter cycle times (50-70s) while maintaining balanced green allocation to further reduce waiting times.
• Keep yellow time at 3-4s and all_red at 2-3s, as these parameters show negligible impact on performance.

Confidence: 75%

Iteration 2

Focus on exploiting the promising region around equal greens of 25s, while exploring asymmetric timings based on flow differences and testing safety parameters (yellow and all-red times) for trade-offs. This balances exploitation of the current best with exploration of parameter interactions and safety constraints.

Analysis (Iteration 2)

The current best configuration with equal green times of 22 seconds for both directions achieves the lowest average waiting time of 68.18 seconds, significantly better than unbalanced or overly long green times. Safety remains intact with no collisions or emergency braking across all tests. Patterns suggest that balanced traffic flow with moderate cycle lengths optimizes efficiency while maintaining throughput.

Key Insights:

• Equal green times around 20-25 seconds provide the best waiting times, with the 22-22 configuration outperforming all others in this batch, indicating balanced traffic allocation reduces delays.
• Increasing green times beyond 30 seconds leads to exponentially higher waiting times, as seen in the 40-40 case with 205 seconds waiting time, due to longer cycle times.
• Shorter yellow times (2 seconds) correlate with better performance in some balanced scenarios, potentially reducing cycle overhead without compromising safety.
• All red time of 2 seconds is consistently used in top performers, suggesting minimal impact on waiting time compared to green time allocation.
• Unbalanced green times (e.g., 30-20 or 15-35) result in higher waiting times on the disadvantaged direction, confirming the importance of balance for junction efficiency.

Recommendations:

• Continue testing variations around the optimal 22-22 green time balance, such as 20-24 or 24-20, to refine the sweet spot.
• Experiment with yellow times of 2-3 seconds and all red times of 1-3 seconds in balanced configurations to assess minor safety and efficiency trade-offs.
• Consider parameter sweeps focusing on green time ratios close to 1:1, with total cycle times under 60 seconds to avoid the degradation seen in longer cycles.

Confidence: 85%

Iteration 3

Focus on exploiting the promising region around the current best (22,22) with fine-tuned variations, while exploring flow-based ratios and slight asymmetries. Also test a safety-related parameter (yellow time) to potentially improve efficiency without compromising safety.

Analysis (Iteration 3)

Analysis of the latest batch shows that symmetric traffic light timings (equal green times for east-west and north-south directions) yield the lowest average waiting times, with the best configuration achieving 50.63 seconds. Increasing cycle times and unequal green distributions generally lead to higher waiting times, while safety metrics remain at zero across all configurations.

Key Insights:

• Symmetric green times (20/20) perform better than asymmetric configurations, suggesting balanced traffic flow reduces delays.
• Longer cycle times correlate with higher waiting times, indicating shorter cycles are preferable for reducing congestion.
• Reducing yellow time from 3 to 2 seconds improved simulation performance (speedup and wall time) but did not reduce waiting time below the symmetric baseline, though safety was maintained.
• All configurations maintain zero collisions and emergency braking, confirming that the tested timing ranges prioritize safety effectively.

Recommendations:

• Focus future experiments on symmetric or nearly symmetric green time configurations with cycle times under 55 seconds.
• Test variations in yellow time (2-4 seconds) to see if shorter times can further reduce waiting without compromising safety.
• Explore adaptive timing based on real-time traffic data if available, but continue parameter sweeps for now.

Confidence: 70%

Iteration 4

Exploiting the promising balanced region around 18-22 for both green times, while exploring safety parameters (yellow and all-red times) that were previously fixed. Also testing flow-proportional adjustments since EW has higher traffic volume, and probing edge cases to better understand parameter interactions and local optima.

Analysis (Iteration 4)

The latest batch shows that unequal green times for east-west (EW) and north-south (NS) directions significantly reduce average waiting times, with the best configuration allocating more green time to EW (23s) compared to NS (17s). All tested configurations maintained perfect safety records with zero collisions and emergency braking incidents, indicating that timing adjustments can improve traffic flow without compromising safety. Longer cycle times and equal green distributions generally resulted in higher waiting times and queue lengths.

Key Insights:

• Unequal green time allocation performs better than equal distribution, suggesting traffic demand asymmetry between EW and NS directions
• Cycle time length has a strong negative correlation with performance - shorter cycles (around 50s) yield better results than longer ones (60s)
• Reducing yellow time from 3s to 2s or all_red time from 2s to 1s provides marginal improvements in waiting times without affecting safety
• Queue length closely tracks waiting time, with the best configurations showing queue lengths around 5.3-5.7
• Throughput remains relatively stable (315-322 vehicles) across configurations, indicating that waiting time optimization primarily affects delay rather than total capacity

Recommendations:

• Continue parameter sweeps focusing on EW/NS green time ratios around 23/17, exploring values from 20-25s for EW and 15-20s for NS
• Test shorter yellow times (1-2s) and all_red times (1s) in combination with optimal green ratios to see if safety can be maintained with further delay reductions
• Consider exploring cycle times between 45-55s to find the optimal balance between flow efficiency and minimal delays

Confidence: 85%

Iteration 5

Focus on refining the asymmetric green time strategy around the best result (ew=23, ns=17), testing slight variations to exploit the promising region. Include exploration of balanced times and parameter interactions (yellow, all_red) for safety considerations. Avoid extreme values that previously showed high waiting times.

Analysis (Iteration 5)

The latest batch of traffic light configurations shows average waiting times ranging from 44.7 to 63.5 seconds, with no safety incidents (collisions or emergency braking). The best configuration from previous experiments (23s EW green, 17s NS green, 3s yellow, 2s all-red) achieves 43.345s average waiting time, and current results suggest favoring EW direction slightly reduces delays while maintaining safety.

Key Insights:

• Increasing yellow time from 2s to 4s significantly increases waiting time (e.g., from 47.95s to 63.525s with same green times), indicating shorter yellow phases are preferable for minimizing delays.
• Equal green time allocation (20s each) results in higher waiting times (50.63s) compared to favoring EW (23/17 at 44.7s), suggesting traffic volume may be higher in EW direction.
• All-red time variations (1-3s) show minimal impact on waiting time when other parameters are similar, but longer all-red increases cycle time and queue lengths.
• No collisions or emergency braking occurred across all configurations, indicating current timing maintains safety even with optimizations for speed.
• Total throughput remains consistent (~320 vehicles), suggesting cycle time efficiency rather than absolute capacity is key to reducing waiting times.

Recommendations:

• Focus next parameter sweeps on EW green times 22-26s and NS green times 14-18s, keeping yellow at 2-3s and all-red at 2s.
• Test shorter cycle times (under 50s) by reducing green times proportionally while maintaining safety margins.
• Consider adaptive timing based on real-time traffic detection to further optimize, but validate with additional safety KPIs.

Confidence: 85%

Iteration 6

Focus on refining the promising region around EW 20-26 and NS 14-20, testing slight variations and safety parameter impacts while exploring symmetry and bias inversion

Analysis (Iteration 6)

The latest batch shows that the best average waiting time of 43.345 is achieved with EW green time of 23 seconds and NS green time of 17 seconds, maintaining a cycle time of 50 seconds. Increasing yellow or all-red times generally leads to higher waiting times due to longer cycles, while safety metrics remain at zero across all configurations. Balanced or slightly EW-favored green times perform better than equal splits.

Key Insights:

• Waiting time is minimized when EW green time is slightly higher than NS (around 23/17 ratio), with the lowest values between 43-47 seconds, compared to 50+ when equal or NS-favored.
• Shorter yellow times (e.g., 2 seconds) reduce cycle time and waiting time without compromising safety, as seen in configuration 9 with 47.95 waiting time.
• Increasing yellow or all-red times extends the cycle (up to 52 seconds), directly correlating with higher waiting times (up to 63.5 seconds) and lower throughput.
• Safety remains perfect (zero collisions and emergency braking) across all tested configurations, indicating current timings don't create unsafe conditions.

Recommendations:

• Explore configurations with EW green time between 22-25 seconds and NS between 15-18 seconds, keeping yellow at 2-3 seconds and all-red at 2 seconds to maintain short cycles.
• Test even shorter yellow times (1-2 seconds) if safety allows, as it significantly reduces cycle time and waiting without visible safety risks.
• Avoid equal green time distributions, as they consistently show higher waiting times (e.g., 50.63 for 20/20 split).

Confidence: 85%

Iteration 7

Focus on refining the promising region around the current best (23/17) by testing slight variations in both directions, exploring slightly shorter cycle times, and investigating the optimal EW/NS ratio that matches traffic flow patterns. Include one balanced test case to confirm equal times remain suboptimal.