.
Operational performance. The experiments demonstrate that the DP logic results in signi.-cantly fewer alerts than TCAS while improving safety. In addition, the DP logic issues fewer reversals. For the cost settings used in this report, the DP logic would strengthen its advi-sories more frequently than TCAS. However, strengthening is a much less severe maneuver than reversing. The cost parameters can be adjusted to reduce the rate of strengthening, but it will be at the expense of additional alerts, additional reversals, or reduced safety. Deciding how to balance these operational considerations will require engaging the TCAS development community.

.
Onboard computation. Since the cost tables and the entry distributions are computed o.ine, very little computation is required on the aircraft. The onboard computation primarily in-volves state estimation, table lookups, and interpolation. Many other approaches to collision avoidance suggested in the literature require substantial online computation, which makes them infeasible for a system that needs to update its decisions at least once per second.

.
Coordination. This report discussed several coordination strategies to reduce the risk of induced collision with other aircraft equipped with collision avoidance systems. Some of the coordination strategies are compatible with the existing coordination mechanism used by TCAS. Further work will be required to re.ne these strategies, but they have already been demonstrated to outperform the current version of TCAS.

.
Unequipped aircraft. Most of this report has focused on encounters with aircraft that are not equipped with a collision avoidance system. It was found that the DP logic signi.cantly outperforms the existing TCAS logic according to both safety and operational metrics.

.
Interoperability. The introduction of a new collision avoidance system will require, at least initially, the ability to interoperate with legacy systems. Many of the coordination strategies suggested in this report are directly compatible with the current version of TCAS. Further work is required to analyze the performance of the DP logic in encounters with TCAS.

.
Surveillance compatibility. The DP logic can accommodate di.erent surveillance technolo-gies. The expected cost tables are not dependent on the sensor error model. Sensor error is accounted for during execution. Incorporating a new sensor system would require updating the tracker, which produces a probability distribution over the state space based on sensor measurements. The rest of the logic that averages the expected cost over the state distribution remains the same. Of course, not all surveillance systems provide the same level of accuracy. If a surveillance system is introduced that provides relatively poor state estimates, the cost of alerting should be decreased so that the safety level is not degraded. This report has focused on the current TCAS surveillance system, but a future collision avoidance system is likely to incorporate improved sensors that reduce the error in bearing measurement, leading to additional performance gains.

.
Aircraft performance constraints. The accommodation of aircraft performance constraints was not a focus of this report, although DP is well suited for incorporation of such constraints. The only performance constraint in the experiments was that the vertical rate be kept within ±2500ft/min, but that range could have just as easily been something di.erent to re.ect the limits of less capable aircraft. One attractive aspect of the way DP handles these constraints is that it does not simply restrict the current space of available actions; it actually chooses the current action with the knowledge that in the future the space of available actions will be restricted. Aircraft with di.erent categories of capabilities would require di.erent cost tables. If aircraft broadcast their performance limits, then these limitations can be taken into account when modeling the intruder aircraft. A collection of di.erent cost tables can be optimized o.ine for di.erent intruder capabilities, and the appropriate table would be indexed into during .ight.
　
中國航空網 www.k6050.com
航空翻譯 www.aviation.cn
本文鏈接地址：Robust Airborne Collision Avoidance through Dynamic Programm(48)