Chapter 16: In Retrospect...

My work with animals had an inauspicious beginning. Early in my college days at the University of Wisconsin, I worked part time cleaning monkey cages and quarters at the school's Primate Research Laboratory. But I soon took on more appealing tasks: I tested monkeys on a variety of learning experiments for graduate students, and did some of their statistical analyses.

My interest in animal learning was enhanced with courses which taught me the basic principles of behavior. In laboratory sessions, I had the opportunity to control, as well as predict and interpret behavior of rat and pigeon subjects.

After I got my master's degree, I was hired by an Army Research Laboratory at Aberdeen Proving Ground, Maryland as a Research Psychologist. There, I developed a method and apparatus for testing the hearing of monkeys, and studied the short term effects of noise on the monkey's hearing.

Several years later, I transferred to another Army Research Laboratory at APG, where I set up a small animal research section in their Biological Sciences Branch. I had the freedom to explore the potential for using animal systems by the military. It was the Vietnam period.

Bird Studies: The first project that I undertook was a study of directional control of bird flight. Two findings were especially interesting: It seemed clear that the bird can be conditioned to follow selected compass direction,1 and that a bird can learn to follow a terrain feature by imitating another bird.2

In the former finding, we began by giving birds no choice but to fly a selected compass direction on a number of training sessions. A test was then done to determine what would happen if they were given a choice to fly in any direction. For the test, the birds were taken to a place that they'd never been before. In the middle of a crossroad, they were tossed skyward. In a representative instance, when the birds received only westward-flight training, they chose to fly west when they had a choice.

The latter finding looked like a case of modeling by one bird on another. The tale begins when the birds were in the early stage of directional flight on a stretch of dirt road. The birds were launched, one at a time, to fly about 100 feet to a work station, where they could peck a key for some grains of food. One bird happened to fly away, and was not seen again for about six weeks—by which time other birds were already flying distances of up to one-half mile on strange roads. The long-lost bird came and perched on the roof of the aviary truck. When the next bird was released, the vagabond voluntarily joined it in flight and followed it all the way to the work station. Then, on future flights, we paired the vagabond with the same trained bird. In a few days, we occasionally observed the vagabond leading the other bird during the flight, which strongly indicated that he had learned to use the road as a directional cue. After a week of paired flights, we sent the vagabond solo; It made it to the work station a half-mile away without hesitation. The phenomenon raised the exciting prospect that one animal could be trained by another.

We were getting one-half mile of reliable controlled flight, when our study was terminated in favor of more urgent work with dogs.

Dog Studies: I was about to begin a new adventure in dog-behavior management. In the mid-'60s, Viet Cong surprise firing devices, trip wires and tunnel openings were being encountered all too frequently by our troops. The U. S. Military Command in Vietnam asked our laboratory to determine if dogs could be trained reliably to detect mines, trip wires, and tunnel openings.

Mine dogs were trained by the U.S. Army during World War II, but we couldn't find any useful documentation that showed how such a dog was trained, or how well they had performed. One document revealed that Army-trained dogs were taught to fear the objects: The dogs experienced an electric shock whenever they came in close contact with such an object. Supposedly, the dogs would thereafter show a fear response whenever they came close to one, which the handler would interpret as a positive indication of its presence. There is no evidence that these dogs were ever used in the field.

The British used another approach that apparently worked. The training information was sketchy, but seemed to be as follows: In the early training trials, visible mines were laid out on a trail, and a food tidbit was placed on top of each mine. The dogs were soon running to mines that they could see from a distance. In another step, the dogs were required to sit at non-baited mines, then were given the food tidbit. Following that, the mines were gradually hidden from sight. It was thought that trained dogs were finding concealed mines by responding to disturbed earth and vegetation which covered the mine, rather than to the mine itself. The work looked encouraging. We took on the task to determine feasibility.

Training Mine, Trip-wire and Tunnel Dogs

We thought that the dog-training methods that prevailed at the time would cause problems for us. These methods produced dogs that were strongly handler-oriented. In our planned work, we needed dogs that would fully attend to the search field, not to the handler. Coercion was a significant feature of contemporary dog training: we questioned whether it would have produced a valid indicator of a dog's willingness to work. We needed the dogs to perform because they wanted to, not because the handler wanted them to.

We took a new approach to dog training. The ground rules for all of our In-house dog work were these: All training was to be done off leash; the dogs were not to be shown how to make a correct response, or be touched, or talked to, or petted and praised during the training session; the dogs were not to be corrected for any errors, or penalized for not responding.3 Correct responses were to be reinforced, and followed by a food tidbit. This approach is the Free-response Method.4

We began our experimental mine and Trip-wire detection work in a small laboratory room. The well-controlled environment permitted us to run many trials in a short period of time. We often got 100 to 120 reinforced responses per day per dog. All training was done off-leash, so as not to influence or interfere with the dog's performance. Positive reinforcement procedures were used exclusively—the dog received a food tidbit for every correct response, and a grand reward on the last response of every short training session.

The dogs were first taught to make a nose-touch of the "mine" object, and were taught ultimately to sit within 2 feet of it. The sit response was the single indicator used when the detection system was completed.

We weren't interested in training the dog to detect particular mine signatures. Mines come in varied shapes, sizes, odors and functions. Some are also handmade, and therefore unique. We opted at first to use mine simulations in the dogs' training. These were small catchall mesh bags that contained pieces of insulated wire, broken parts of real mines, dabs of heavy grease, rusty nails, pieces of previously worn clothing and other handy objects.

The dogs learned the nose touch response quickly in two to three "shaping" sessions. In these sessions,5 the "mine" bag was positioned on the floor between the handler and the dog. In later sessions, the mine was tossed from one position to another within the room. The dog would run to the mine and touch it with his nose. It was almost like a "fetch" game, except that the dog never retrieved the object. Next, the sit response was linked to the nose touch response to form a more complex performance: At the moment the nose touch was made, the dog was told to sit, and then helped to make the response. After several more sessions, we were able to discontinue the command to sit and other assistance. The dog sat automatically following his nose touch.

The dog was then trained outdoors. The mines were placed at varied intervals along a narrow, long dirt road. The dog would run to a mine position, nose-touch the mine, then sit in front of it., In several training stages, the mines were gradually concealed until they were out of sight. At each stage of concealment, the earth was also disturbed at other non-mine-emplaced positions. The mines were concealed with dirt and/or natural vegetation. Other non-mine positions likewise had dirt and/or vegetation disturbance. The dogs were allowed to smell the non-mine positions for a moment, but were then urged to move on.

When the dogs became very good at finding concealed simulated mines, we began concealing varied single objects, i.e., empty milk carton, glove, newspaper, piece of rubber hose, soiled clothing, small antipersonnel mines (minus detonators), etc. The strategy that we had in mind for the dog to learn was to find disturbed earth and vegetation positions; then to closely sniff those places to determine if anything was concealed there—anything. If the dog detected something that was different from the surrounding medium, he was supposed to sit close to it. If nothing different was detected, he was to move on in further search.

We found that when mines were emplaced a short distance off the trail or road (usually 20-30 feet), the dogs were able to find them by following the track of the person who laid it.

We also discovered that dogs detected trip wires by sight. The off-leash dog's normal working speed was 2-3 miles per hour. Whenever the dog came across a human ground scent, he stopped short and started searching for a trip wire or a mine emplacement. Some dogs bobbed their heads up and down several times. I tried bobbing my head at the dog's level and found that even the thinnest nylon filament could be seen that way. The trip wire could be made to move against the horizon and other backgrounds (what intelligence). If the dog detected a wire, it then went to the wire and sat facing it.

At that point of our work, we called for proposals from contractors to train operational Mine and Tunnel Dogs, and to train a platoon of Army dog handlers to operate and maintain the system.6

Training Explosive Detector Dogs

I stopped the in-house work on the Mine and Tunnel Dog, but continued to monitor the work of the contractor. My next project was to find out if dogs could reliably be used as explosives detectors. This was still the Vietnam War period, when some individuals and groups protested in this country with explosive devices. Training Explosives Dogs was a wholly different problem from the previous one. In this one, the dogs would have to recognize the signature odors of explosives. We chose to train our in-house dogs on dynamite and black powder.

By now, our laboratory had several test-sophisticated dogs with which to continue our new work. The dogs were already under good basic control. We needed these dogs to learn a new function and work strategy. We again began with a nose-touch response, but this time, we required the dog to hold the nose touch for about two seconds—the two-second duration was gradually shaped. Next, we taught the dog to do a two-position discrimination. The objects used were open Mason© jars. One jar of each pair contained a bit of dynamite wrapped in gauze; the other jar contained only the wrapped gauze.

The two-position discrimination procedure: I sat on a stool facing the dog. The discrimination "setup" was between us. I took two bottles from a bottle rack that was at my side, and placed them in a bottle holder (to keep the bottles from tipping over) in front of me. When the dog's nose was at the opening of the positive bottle, I said GOOD, then gave the dog a food tidbit.7 We played the shell game with the jars. That is, after each trial, the bottles were raised from the holder; then they were either switched in position or put back in the same holes. After the dog had four such exposures, the bottles were replaced with a clean pair. Many bottles were washed each day.

Next, the dogs were taught to move out in search. The initial work began with two small cardboard boxes—one held a bit of explosive. The boxes were placed two feet apart, and were lined along a wall. All trials started 20 feet from the first box.

To begin a trial, the dogs knew to come to heel first. The dogs were taught to do this automatically without being asked. The dog stood at my left until it got the signal to go out in search. The send-away signal was a light touch to the back of the head. The dog went and sniffed each box. When the dog had his nose at the target container for at least two seconds, I said GOOD, went up to the dog and gave him a tidbit.

When the proper search behavior was established, ten matching cardboard containers were lined up along the wall for every search try. Only one container held the explosive. The single detection indicator used in this training step was a sit response when made close to the target container with the explosive. After the dog sat, I walked up to the dog, stood still a moment, then said GOOD, and gave the dog a tasty tidbit. On the last trial of the session, the dog was told, OUT, and a grand reward given. It was a shell game again—all boxes were picked up and/or touched and the positive box was positioned in another place, ready for another try. After two or three days, the boxes were destroyed and replaced by another set.

The dogs learned the signature odor of the second explosive, black powder, by associating it with dynamite. The two explosives were wrapped together in gauze and placed in the positive bottle.8

When our dogs reliably detected explosives in a more-realistic setting—the positive and negative boxes were hidden in a military warehouse—a call was made for proposals to develop a Explosives Detecting Dog System to assist the police.9 Dogs were successfully trained at the University of Mississippi, and tested at our laboratory before they were given to the New York City Police Department. Apparently, these were the first working explosives (bomb) dogs in the country.10

Training a Heroin-detecting Dog

Following another detection inquiry from a U. S. Military Unit in Vietnam—this time about a dog's ability to detect heroin—we began work immediately on the problem. This would be a more difficult detection problem for the dogs, because pure heroin has little odor. We used the same test-trained dogs in this work.

We discovered that acetic acid is a byproduct of heroin manufacture, and is often found in very dilute amounts with the drug. The training strategy was to get the dog to detect acetic acid first, because its odor can be made readily detectable. Then the acetic acid would be paired with "pure" heroin, followed by gradually "fading" the acetic acid out of detection range. At which time, the dog should be detecting the weak-odor heroin itself.

We used the same 2-position discrimination procedure, with Mason® jars holding the test substances. One jar held an ounce of weak acetic acid solution; the other, an ounce of plain water. After the dog responded consistently to the acetic acid, the solution was gradually reduced in strength until the dog had difficulty in detecting the correct jar. Then, for the next training step, the strength was raised just enough, so that the dog readily detected it.

A small amount of pure heroin was placed in a test tube and set in the Mason® jar containing one ounce of the weak, but detectable acetic acid. The other jar contained one ounce of water, and a test tube that contained a small amount of pure white talc. In a series of trials and sessions, the acetic acid was made progressively weaker, until finally it was removed entirely. Water was then removed from both jars. In this setup, the dogs continued to respond to the jar that contained heroin.

It's probable that the "pure" heroin had just enough acetic acid as a byproduct of the heroin's manufacture, that the dogs were responding to that substance rather than the heroin. We did one more step, to find out what would happen if we put the acetic acid in the "negative" jar. In which case, the dog should stop responding to just the acetic acid. In that step, the positive jar held one ounce of water and a small amount of heroin in a test tube; the negative jar held one ounce of weak acetic acid and a small amount of talc in a test tube. Over a series of trials, the acetic acid was varied in strength in the negative jar from weak to undetectable. Responses to the negative jars were soon ignored.

Eventually, the dogs chose the heroin-containing jar consistently. Unquestionably, we showed that dogs can detect the weak heroin substance. One month from the time we received the query, we described the procedure to the Military Unit in Vietnam that requested the information. We were not able to do a follow-up on how they used the information. Although we did not test our dogs under real world conditions, I feel that we had the first valid Heroin-detecting dogs.

Controlled Aggression

On occasion, I spent time with military dog-training units at other Army bases. One such time, while doing experimental work on one of my projects, I had the chance to observe guard-dog training. One particular dog caught my eye. No matter what the trainer did, the dog would not grab a sleeve or play tug-of-war with a towel. Otherwise he was outgoing, friendly and appeared fearless. I asked the chief trainer if I could work with that dog for a couple of weeks. He not only consented, but also lent me a person to help me.

The work was set up in a large empty warehouse. For the training trials, the harnessed dog was tied on a 4 foot leash to a wall fixture. The dog faced a large warehouse door that was about 60 feet away. A food pan, containing one tablespoon of a tasty food, was placed about midway between the dog and the door opening.

In the first training step, the dog learned the meaning of the sound, OUT. After saying the sound in a moderately loud voice, I ran to the pan, picked it up, and placed it in front of the dog. After he ate the food, I placed another dollop of food into the pan, and set it down away from the dog. The dog got six such trials a day for three days. The interval between trials varied from 30 seconds to two minutes. At the end of the session, the dog was led out of the warehouse, and handed to the helper. The dog was then taken on a pleasure stroll, usually through a wooded path to a nearby Bay where he usually played in the water.

On the fourth day, we began target training. The lights in the warehouse were turned off for the remaining trials of the experiment. The only light came from the doorway. The helper became the target on every trial. The trial began when the target appeared in the doorway and walked to the food pan in a stealth manner. I positioned myself close to the dog. When I saw the target, I bent close to the dog, and repeated the sound, WATCH IT, in a low hushed tone of voice. Every five feet, the target stopped for a moment, before coming on again. When the target was about ten feet from the pan, I said, OUT. At the sound, the target ran out of the building. We gave six trials of this procedure daily for two days. The dog observed the proceedings with no overt emotion.

On the sixth working day, the 6-trial session was the same, except that the target now reached the food pan, seized it, and ran off with it. I then ran after the target; the target quickly laid the pan down near the open door, and continued running out of sight. I retrieved the pan, set it down in the usual place, and after a brief wait, I said the sound, OUT. Then I placed the pan in front of the dog. Still, no unusual reaction from the dog.

On the seventh day, two trials were given. The first trial was done without the target. It was the same procedure as on the first three training days. On the second trial, the procedure of the sixth day was done with this change: This time, the target ran off with the food dish, and the dish was not retrieved. On this, and later trials, I no longer chased the target. I just waited another several minutes near the dog without talking to him. The dog was then released and brought out of the warehouse and handed to the target-helper. At this point in the experiment, it was up to the dog to do something that would prevent the target from running off with his food.

On the eight day, one trial was given. The "target-present" procedure of the previous day was run. Still, no unusual reaction. On this particular day, the dog also did not get his end-of-day food ration.

On the ninth day, the dog uttered a brief bark when he saw the target. When the target heard the sound, he turned and ran off without the food dish. I immediately said, OUT. The dog then got the food. Whenever the dog failed to react to the target, the training session was terminated, and the dog lost his chance for food for that session.

On subsequent days, the barking response was strengthened. By "shaping," we finally got an aggressive temper to the response. Whenever the dog reacted consistently at a particular temper level, we raised the ante: The dog was then required to be more expressive in his reaction, if he didn't want to lose the food.

Finally, we were able to get a combined growl and snarl, to the sound, WATCH IT. We could get that reaction even when the target was still out of sight.

After a month of training, we demonstrated the dog to the guard-dog group. Before the demonstration started, anyone could come and pet the dog. The dog appeared to be an average friendly pet. On the trial, a single person was told to approach the dog in a suspicious manner, from a distance of 50 feet. When the person was about 15 feet from the dog, I said, WATCH IT, in a distrustful tone of voice. The dog let out an impressive-sounding vocal response. The person stopped in his tracks. About five seconds later, I said, OUT. The dog's vocalizing ceased immediately. The target-person then came to the dog and petted him. He admitted that the dog's vocal response was pretty scary.

The experiment raised a lot of questions, among which were, "Is the emotional response real or sham?" or, "Would it be safe for a policeman-handler to have such a dog living with his family?" or, "Would this otherwise friendly and outgoing dog physically attack a person?" Anyway, the dog was put back into the program, and I continued doing other work.

Control of Dog Movement in Open Field

One of our more interesting projects was to get the dog to move from point A to point B without his handler—by telling the dog via radio signal to make a correction in his direction of movement, whenever one was needed (see Computer-assisted Training, next).

One of the things that we were able to do was get a dog to move by himself down a road for about one-half mile, until he came to a choice point, where he was given a change-of-direction signal. If the choice point was a crossroad, we could get the dog to turn left, right, go straight ahead or go again in the direction in which he came. The trained dogs very seldom made an error at crossroads. If the working dog was in open space, and came up to a brush-line, a road, trail or structure, he could be directed to move in any direction, relative to the feature.

The dog's movement in space was under continued terrain control. That is, if the dog was on a trail, he would continue to follow the trail, until the radio signal told the dog to shift his direction of movement to another close-by terrain feature. We encountered one problem that was not resolved by the time we terminated the work, namely: If the dog was in a large open field, we could not get the dog to go in a particular direction. That is, we were unable to get the dog to latch on to a significant distant terrain feature. The solution was to let the dog move in any direction, until he came close to a terrain feature that had directional choices. From there, the dog would be moved to another terrain feature that was closer to the direction we had in mind.

Computer-assisted Training

Actually, we didn't have a computer (it was the early 70s), but we had programming equipment that worked like one. First, we made a flow chart of every contingency that could happen on any particular training trial in Directional Training. Then, logic modules in solid-state programming equipment were wire-connected to run most of the early trials automatically, without trainer assistance.

Directional Training. The dog began the work in a box that measured 4 feet by 8 feet by 6 feet high. Inside the box, the dog made key-press responses with his nose. Three keys were available for pressing. A home key was positioned midway along the 8-foot side. Work keys and feeders were positioned at each of the two ends. After a series of preparatory training steps, the dog went to whichever work key that was lit. For each key press, a tasty food pellet dropped into a cup below the key. After a random number of key presses (between 2 and 6), the work-key light turned off, and the feeder locked up. At which time, the home-key light turned on. The dog would then go to the home-key and press it. This turned off the home-key light, and turned on one of the end work-lights. The dog went to the lit work-key and pressed it for food.

The home key simply reset the equipment for another trial (it had no feeder attached). It also put the dog in mid-box position to begin another trial.

In additional training steps, without going into detail, we got the dog to use sound, rather than a lighted key, to tell him which key could be worked for food. It worked this way: After resetting the equipment with a press of the home key—the home-key light still turned on and off, appropriately—the dog could choose to go toward either work-end. If he happened to go toward the wrong end, a tone came on and stayed on until the dog changed course and entered the correct end. After a set number of food-rewarded key presses, the feeder locked up and the home key light turned on.

The trainer had little to do in this automated setup. He put the dog into the work cage; pressed the start button on the equipment; then he could go get a cup of coffee, while the dog worked. A recorder gathered continuous data. After a preset number of trials, the equipment shut itself off; The dog was then removed from the cage.

The next apparatus was Y-shaped. Each leg was and 16 feet long and 4 feet wide. The sides were wire mesh, so that the dog could be seen at all times. The home key was at the bottom leg of the Y. Work keys and feeders were found at the upper ends of the other two legs. The choice point was where the legs joined. After pressing the home key, the dog moved toward the work ends. Most dogs favored going into the right leg. When the dog entered the correct leg, no sound came on—the dog continued to the work-end of the leg, and started key pressing for the food tidbits. Whenever the dog entered the wrong leg at the choice point, a tone came on. When a sophisticated working dog turned on the tone (by stepping into the wrong leg), he immediately changed course without pausing or missing a step. At this and the next stage of the experiment, the trainer controlled tone onset and offset.

The last training apparatus was cross-shaped. Each leg was 16 feet long. A home key was at the end of one leg; work keys and feeders were at the ends of the other three legs. The dog was trained to go straight through the choice point, without turning, if no tone came on. When the procedure called for a turn to be made into one of the two side legs, the trainer turned on the sound at the choice point. At the sound, the dog began turning and kept turning until the tone came off. The trainer manually turned off the tone when the dog was aimed in the correct direction; the dog then stopped turning and went into the correct leg. At first, we had problems getting the dog to go in a particular direction when the tone turned off. We soon learned to turn the sound off when not when the dog's body was correctly facing, but when the dog's head faced in the correct direction. Learning significantly accelerated when we made the change.

In the next and last training phase, the work was done in a variety of settings in unrestricted space. Among other things, this work showed that some specialized dog training can be more efficiently and effectively assisted with a computer-managed program.

Mentally-Retarded Dogs

A senior military dog-Instructor told me that he had two mentally retarded dogs—both were from the same litter. He asked me if I could learn something more about their learning ability.

At the time, I was doing the first phase of the directional-control work. I decided to put these two dogs through the same initial program. I stopped their work when they were able to change direction to tone almost as well as "normal" dogs, in the 4 feet by 8 feet box. However, the data clearly showed a diminished learning capacity in both dogs: These dogs were given 5 to 10 times more trials to reach a set performance criterion at each training step than were given the average run-of-the-mill dog. This work suggests that a simple, reliable training procedure could be used to test the mental capacity in dogs.

Police Dog and Sheep Dog Training

The purpose of the aforementioned direction-control study was to have the dog move from one ground point to another, with radio signals. A single tone was used to get the dog to shift from one terrain feature to another, until he reached his destination.

The present study was another instance of controlling the off-leash dog in the free field. Police dogs and Sheep dogs are trained to respond reliably to guidance signals from the handler, and to perform a useful function at a distances from the handler. The training involves getting the dogs, by voice or whistle, to move to their right, move to their left, move in, move away, move slow, halt and stand stay, and/or down stay—always in reference to some animate subjects who are away from the handler in the work field. For instance, if you told the dog to move to the right, he would move right in a circular movement around the subject out there until he's told to do something else, also done in reference to the subject that he is facing. The police dog's task, trained this way, would be to bring to a halt, and hold one or more persons at bay, with controlling directions from the officer.

The dog used in this study was already under control in several response components of the system, before this work was begun. The dog already knew the voiced commands to lie down, halt and stand-stay, move toward a position, move slowly or at a steady trot, and recall back to the handler.

Additional commands were learned in the present study, namely: "away" movements relative to the target: HI (sounds like HIeee; the dog moves to his right in a circular movement around the target), and LO (sounded like LOuuu; the dog moves to his left, in a circular movement around the target). If we had formalized the system, the more traditional sheepdog commands would have been used instead. The dog's movements were trained with extensive, but gentle, leash prompting.

Some dogs show no natural interest in this kind of police or sheep work. However, they can still perform reliably with contrived rewards. The dog in this study was trained with food—a contrived, provisional reward. If the work had been continued to completion, a shift in a nonfood incentive would have been attempted. The incidence of dogs who don't come to enjoy this kind of work for its own sake is rare, indeed. It is hoped that the police dog will work to hold a stranger at bay, just for the inherent pleasure that he might derive from doing that.

Since the work had no immediate relevance to my current projects, the project was terminatedafter control of these response components, was obtained. The dog was still on-leash when the work ended.

Olfactometric Studies

We were about to begin an exciting series of experimental studies on the smelling ability of dogs. We had already built a special odor test-chamber for dogs, and we were ready to let a contract for a specially built olfactometer (designed by a consulting service). But the Vietnam War ended and all dog work stopped. I then left Government service.

John J. Romba, author

top ::: home