Following the Herd After Dark
Fieldwork with WNE India in Jharkhand, connecting nighttime elephant tracking and the practical demands of conservation
By Siddharth Mayya and Aditya Palta
Early-warning systems are used across Asia and Africa to mitigate human-elephant conflict by providing advance notice of approaching elephants, allowing communities to avoid dangerous encounters and organize timely responses. These systems range from community volunteers reporting sightings to automated cameras, with alerts distributed through phone calls, text messages, and local networks. However, reliance on manual monitoring places a substantial burden on communities, particularly at night. Research in Wayanad, India, found that all surveyed households undertaking night guarding experienced sleep-related fatigue, underscoring the need for technologies that reduce the burden and risks faced by volunteers and forest staff.
Habitat Robotics has been working with WNE India to evaluate thermal-camera-equipped drones as a means of extending early-warning coverage in fragmented landscapes such as Chandil, Jharkhand, where elephant movements between forest patches, agricultural fields, and settlements are difficult to monitor using fixed cameras alone. Drones can survey larger areas and detect elephants at greater distances, offering a promising complement to ground-based monitoring systems. During fieldwork in Jharkhand in October 2026, Habitat Robotics and WNE India combined daytime aerial surveys with nighttime thermal observations to examine how drones can support elephant monitoring around fields and forest edges.
A stitched aerial view from our daytime survey near Chandil, showing forest, farmland and settlements alongside a highway junction and railway.
On October 2, successive night flights recorded elephants moving through and feeding in paddy fields near Chakulia, Jharkhand. The video on the left shows some of the thermal drone footage captured by our team, allowing them to estimate the number of elephants in the herd as well as their movement and behavior. Automating these detections and enabling the drone to estimate the direction of movement of the elephants requires the use of computer vision algorithms to distinguish the elephants from other thermal objects in the scene such as vehicles, people, and buildings. The video on the right shows an initial experiment combining an open-vocabulary detection model with a segmentation model to delineate elephants in the thermal footage and follow them through the video. We are testing several approaches to examine how reliably elephants can be detected, counted, and tracked under different conditions.
Left: Original thermal drone footage of elephants in paddy fields. Right: A computer vision overlay highlights and follows the elephants across frames.
Accurately estimating the direction of elephant movement from aerial footage requires distinguishing the animals' motion from that of the drone itself. To compensate for camera movement, we applied intensity-based image registration, excluding segmented animal regions when aligning the background across successive frames. The side-by-side comparison shows the original footage on the left and stabilized footage on the right, with short trails illustrating each tracked animal's recent movement in image coordinates and relative to the stabilized background, respectively.
Left: Drone motion can be mistaken for elephant movement. Right: Background stabilization compensates for drone motion, making elephant movement easier to assess.
In landscapes where agricultural fields are interspersed with forest patches, continuous observation is rarely possible. Elephants can disappear beneath dense canopy and emerge elsewhere, creating inevitable gaps in detection. To address this, we are developing uncertainty-aware tracking algorithms that maintain a belief state over the herd’s likely position, allowing the drone to anticipate where elephants may reappear and position itself to reacquire them.
Perception, however, is only one part of building an effective aerial early-warning system. During our field deployments, battery life was a major operational constraint, requiring drones to land approximately every 30 minutes for battery replacement or recharging, creating gaps in monitoring during which elephants could move considerable distances. Sustaining these operations night after night also places significant demands on field teams, who must manually coordinate flights, monitor the animals, and manage batteries in challenging conditions.
Addressing these limitations requires moving beyond individual flights toward more autonomous, persistent monitoring. Deciding where to search, when to follow a herd, and when to return for recharging presents a multi-objective planning problem under uncertainty. These decisions require balancing tracking performance, energy consumption, and operational constraints, while adapting to new observations and changing conditions in the field. Habitat Robotics is bringing together expertise in robot autonomy and field ecology to develop and evaluate these approaches in real-world conditions, with the ultimate goal of translating advances in autonomous monitoring into timely, actionable warnings for communities and response teams. Our longer-term objective is to translate these advances into a reliable 30-minute early-warning window, giving communities and response teams more time to prepare for approaching elephants.
Our focus areas
Detecting and tracking elephants, estimating their movement, and maintaining useful information when visibility is interrupted.
Coordinating where to search, when to follow a herd, and when to return for recharging.
Working with conservation teams to shape deployment, maintenance, and cost-sharing models that sustain these systems over time.